Written and reviewed by Bc. Petr Kučera, MSc., MBA
You can have the most beautiful website from a top designer, but if it runs on a tractor engine, it won’t reach your customers in time. So we set out to bust the myth that “hosting doesn’t matter” and declared a battle for milliseconds: we took one website, made two identical copies, and sent them into a duel on two different servers.
To keep the twins comparable, the website was identical on both sides. Only the hosting differed: the server and its object cache.
We ran the test on 19 March 2026 in PageSpeed Insights, running each variant at least five times and reporting the median. PageSpeed Insights runs its lab test from a Google data center in North America, Europe, or Asia, so server response is measured from there, not from a visitor’s device. PageSpeed Insights showed no real-user data for the test URLs (“No data”), so this is purely a lab result.
The tested WordPress site is no longer live, and ppcone.cz has since moved to different technology. The results apply to that site, not today’s ppcone.cz.
On desktop, the results were similar. But most internet use today is on mobile, where the difference was clear. All numbers in the table are medians from PageSpeed Insights on 19 March 2026.
| Metric | Shared hosting | Our own server | Difference |
|---|---|---|---|
| Mobile performance | 62 points | 95 points | 33 points |
| Desktop performance | 96 points | 98 points | 2 points |
| Time to First Byte (TTFB) | Roughly 940 ms | Less than 50 ms | Roughly 890 ms or more |
| Main content rendering on mobile (LCP) | 5.7 s | 1.2 s | 4.5 s |
Shared hosting scored 96 points on desktop, which could easily make you relax too soon. On mobile, though, the gap was 33 points, and on shared hosting the main content appeared 4.5 s later.
TTFB, or Time to First Byte, is a hidden time thief. It measures the interval from the start of navigation to the first byte of the response. That includes redirects, DNS, connection setup, and TLS, not just the server’s processing time.
The browser must wait for the first byte before it can start downloading referenced images, styles, and scripts. High TTFB adds time to every metric that follows. LCP (Largest Contentful Paint) measures when the largest content element is rendered. Its components are TTFB, the delay before loading that element, the time spent loading it, and the delay before rendering it.
According to web.dev, a TTFB of 0.8 s or less is a rough guideline, and anything above 1.8 s is poor. TTFB is not one of the Core Web Vitals. Shared hosting, at roughly 940 ms, landed between 0.8 and 1.8 s, in the “needs improvement” band. Our own server came in under 50 ms, roughly 19 times faster or more.
The TTFB gap was roughly 890 ms or more; the mobile LCP gap was 4.5 s. Server response explains only part of it. The rest lies in later phases of loading and rendering the main element, so we do not attribute the full 4.5 s to hosting.
On shared hosting, a server’s performance is split among multiple websites. web.dev explicitly names shared hosting as a cause of slow server response, noting that it is generally slower. In our view, a faster CPU, faster disks, or object cache can help, depending on the specific bottleneck.
The hosting made a big difference for this website. That does not mean every dedicated server beats every shared hosting plan.
One bonus we noticed outside the table: the WordPress admin was much faster on our own server, which made everyday work less of a grind.
Having Cloudflare doesn’t mean you can ignore website speed. By default, Cloudflare doesn’t cache HTML. It can cache HTML with the right cache rules or APO. In our test, both sites had identical Cloudflare settings and page cache turned off, and the gap between the two hosting setups remained.
In WordPress, Redis object cache stores application data between requests. It is not a ready-made copy of the whole page. WordPress still builds the page; object cache keeps the data ready. It is one tool among several, not a miracle speedup.
Core Web Vitals are three metrics of user experience. Google assesses them at the 75th percentile of visits, based on data from real users, not a single test.
| Metric | What it stands for | Good value according to Google |
|---|---|---|
| LCP | Largest Contentful Paint: rendering the largest content element | 2.5 s or less |
| INP | Interaction to Next Paint: response to interaction | 200 ms or less |
| CLS | Cumulative Layout Shift: shifts in page layout | 0.1 or less |
| TTFB (not a Core Web Vital) | Time to First Byte: time until the first byte | web.dev guideline: 0.8 s or less |
INP replaced FID on 12 March 2024. For LCP, anything above 2.5 s and up to 4 s needs improvement, and anything above 4 s is poor. Our lab LCP of 5.7 s on shared hosting would fall above the 4 s threshold (poor), while 1.2 s on our own server would fall below 2.5 s (good).
What it means for SEO:
Speed is one signal, but content is the foundation. We cover how content translates into visibility in AI answers in a separate article.
A popular rule says “every extra second cuts conversions by 7%.” We looked at where it comes from: a November 2008 Aberdeen Group survey of more than 160 organizations. That makes it a historical figure from a single survey, not a rule for 2026. The same goes for other frequently cited numbers, so we list each one with its year, author, and study type.
A 1-second delay meant 7% fewer conversions.
A 100 ms delay was associated with fewer conversions: down 7.1% on mobile, 2.4% on desktop.
Mobile bounce rate: 14.1% (0.7 s load), nearly 29% (2.7 s).
53% of mobile site visits were abandoned when loading took more than 3 s.
Model: increasing load time from 1 to 7 seconds raises bounce probability by 113%.
A 0.1-second improvement in each of the four speed metrics tracked was associated with higher conversions: 8.4% in retail and 10.1% in travel.
Landing page LCP improved by 31%; sales rose by 8%.
A B2B site loading in 1 s had a conversion rate three times that of one at 5 s.
The share of stores with good LCP rose from 57% to 96%, and the share passing Core Web Vitals from 48% to 72%. Mobile conversion rate from organic search rose by 8.9%.
Revenue per visitor up 53.37%, conversion rate up 33.13%.
According to the authors, Amazon experiments found a 1% revenue drop with a 100 ms delay.
How to read the numbers:
Still, the studies point the same way: faster loading, especially on mobile, goes hand in hand with better results.
If you pay for ads in Google Ads, Sklik (the ad platform of Seznam, the Czech search engine), or Meta, speed matters beyond SEO. Google Ads counts a click even if the visitor never reaches your site. With a mobile LCP of 5.7 s, as in our shared hosting variant, you may be paying for clicks from people who never see your offer. We explain how targeting, the auction, and the landing page work together in our article on how PPC campaigns work.
Speed is one part of landing page experience. It is not a pricing formula.
In Sklik, the Quality column is informational only and has no direct effect on ad position. Position depends mainly on your maximum cost per click and the quality coefficient. Sklik’s help pages don’t mention landing page speed and don’t describe any direct link between speed and click price. But if a visitor leaves before seeing the page, the paid click is wasted. We cover advertising on Seznam in a separate article.
With Meta, the main link to speed is analytics. Meta recommends placing the pixel code in the page header so it fires earlier and has a better chance of recording the visit before the person leaves. On a slow site, some visits may be missing from your data. We cover adding server-side tracking in a separate article.
80 client reviewscf-cache-status. HIT means the page came from Cloudflare’s cache. MISS and DYNAMIC mean your server had to send it. Without cache rules or APO, Cloudflare doesn’t cache HTML by default.Based on our experience with this test, we’d consider four factors when choosing a hosting provider. They aren’t a universal law; they’re simply what set our faster server apart.
Before you switch hosting, measure where the bottleneck is. If it’s the server, more powerful hosting can help. If it’s the website itself, the site needs work. We can help with that as part of our web development and other services.
Hosting affects speed mainly through server response. On the same website, TTFB was roughly 940 ms on shared hosting and under 50 ms on our own server. In the lab test (median of at least five runs), the mobile score was 62 vs. 95 points and LCP 5.7 vs. 1.2 s. Server response explains only part of the LCP gap. On shared hosting, performance is split among websites. TTFB also includes redirects, DNS, connection setup, and TLS, so not every slow TTFB is the host’s fault.
According to web.dev, a rough guideline is 0.8 s or less, and anything above 1.8 s is poor. TTFB isn’t a Core Web Vital, but it adds time to every metric that follows.
Google says its ranking systems use Core Web Vitals. Good Core Web Vitals values don’t guarantee high rankings. Content quality matters more than page experience. According to Google, chasing a perfect score purely for SEO may not be time well spent.
Lighthouse tests mobile and desktop under different conditions and scores them differently, so don’t compare the two scores directly. In our test, shared hosting scored 96 on desktop but only 62 on mobile; our own server scored 98 and 95.
In Google Ads, it can, indirectly, but it isn’t a rule. Quality Score is diagnostic. Landing page experience is both a Quality Score component and a factor in the auction, and Google says higher-quality ads can often lead to a lower cost per click. Faster loading is among Google’s recommendations for improving landing page experience, but Google hasn’t published any fixed formula linking speed to click price.
Sklik’s help pages don’t mention speed. The Quality column is informational only, with no direct effect on position, which depends mainly on maximum cost per click and the quality coefficient.
The test showed that even an optimized website can be held back by its server. Same code, same images, same Cloudflare, yet mobile scored 62 vs. 95 points, and the main content appeared 4.5 s later on shared hosting. On desktop, everything seemed fine.
If you’re serious about doing business online and running paid ads, a desktop score isn’t enough. Measure mobile, TTFB, and LCP, and find out where loading gets held up.
We can run a similar analysis of your website, pinpoint what slows its loading, and show what it may mean for your ads and SEO.
About the author
Bc. Petr Kučera, MSc., MBA
AI Systems & Marketing Strategist, founder of PPC ONE
Petr discovered the power of performance campaigns back in 2011, and they have been an inseparable part of his professional life ever since. His edge comes from working with more than a hundred companies.
Facts verified as of 02/10/2026.