We Put 10 Real Sites on a Throttled Phone. Here’s Who’s Actually Fast.

Most platform speed comparisons test demo sites, on fast connections, published by whoever sells the winner. We did the opposite: ten real, live small-business websites, one deliberately miserable phone profile, and a commitment to publish whatever came out — including the rows where we lose. Spoiler: there are rows where we lose.

The method, so you can check our work

On 2026-07-12 we loaded each site’s homepage in headless Chromium (driven by Playwright), emulating a 390×844 phone. The network was throttled to roughly Slow 4G — 1.6 Mbps down, 750 kbps up, 150 ms of round-trip latency — and the CPU slowed 4×, because your customers’ phones are not your developer’s laptop. Each homepage was loaded three times and we report the median. We captured time to first byte (TTFB), First Contentful Paint (FCP), Largest Contentful Paint (LCP), Cumulative Layout Shift (CLS), transferred bytes, and request count, through the load event plus a 3.2-second settle window.

The field: three StoryPress-built sites (storypress.app — yes, our own homepage — make3.co, and auresources.io), two Wix sites (crazyaboutcookies.com and ilovebakedny.com), two Squarespace sites (tandemcoffee.com and saltwaterbakeshop.com), two WordPress sites (pinchofyum.com and thekitchenwhisperer.net), and one Webflow site (lattice.com).

Caveats, stated plainly:

• This is a lab test, not field data. Google’s CrUX dataset measures real visitors; we measured one synthetic visitor with identical weather for every site. Lab numbers rank; field numbers convict.

• Ten sites is a probe, not a census. Draw directional conclusions, not verdicts.

• One Wix site (ilovebakedny.com) failed DNS resolution on all three runs and was dropped. A site that doesn’t resolve is admittedly very fast to give up on, but we didn’t count it.

• One pinchofyum.com run timed out at 90 seconds; its median covers the two runs that finished.

• Both WordPress sites run heavy ad stacks; their request counts include ad-tech polling during the settle window, and the table flags them accordingly.

What won — and what it cost us to print this

The outright winner is a Wix site. crazyaboutcookies.com swept every column: a 444 ms LCP, 0.01 CLS, 307 KB transferred, nine requests total. Nine. It won the way anything wins on a slow connection — by not sending things. Before anyone updates a pitch deck: this is one very lean page, not a platform coronation. But a result is a result, and it’s the best one in the table.

The worst result is just as instructive. saltwaterbakeshop.com (Squarespace) shipped a 4.3 MB homepage over a 1.6 Mbps line and posted a 12.8-second LCP. That’s not bad luck; that’s division.

Squarespace deserves credit where it’s due: it posted the best TTFBs in the entire test — 66 and 73 ms — beating every StoryPress site. Their servers answer fast. Then the payload arrives. Fast waiter, heavy tray.

The two WordPress sites landed mid-pack on LCP (3.6 and 3.8 seconds) with the heaviest payloads in the test — 4.7 and 6.1 MB — and request counts in the thousands, courtesy of ad tech running laps during our settle window. Neither result embarrasses WordPress the software; both show what a monetized WordPress site typically carries around.

Webflow’s lattice.com had the strangest profile: nothing painted for 3.3 seconds, and then the largest element appeared at the same instant as the first one. A blank stare, then the whole answer at once.

And now our own report card. Two StoryPress sites behaved the way the architecture is supposed to: auresources.io painted its largest element in 664 ms and make3.co in 1.5 s — the second- and third-fastest LCPs in the test. Then there’s storypress.app: a 5.2-second LCP, 1.4 MB, 114 requests. Our own homepage got out-painted by both WordPress food blogs. We measured it, we’re printing it, and the parties responsible (us) have been notified (by us).

Two more warts, since we’re counting. auresources.io logged a 0.91 CLS — a superb first paint followed by the furniture sliding around the room. Anything above 0.25 rates as “poor” on . And our TTFBs (248–326 ms) were merely mid-pack; edge hosting helps, but it isn’t a monopoly. Squarespace has edges too.

The actual lesson of the table

Rank the nine sites by page weight and you nearly reproduce the LCP ranking. The winning page carried 307 KB; the losing page carried fourteen times that. On a throttled connection, weight is destiny and the platform logo is decoration.

The field data mostly agrees

Our probe rhymes with the population-scale numbers. The CrUX-based Core Web Vitals Technology Report puts Wix at 74.86% of origins passing all three Core Web Vitals and Squarespace at 70.39%, versus WordPress at 46.28% (). Managed platforms mostly ship a fast floor. WordPress ships a warehouse of parts; the floor depends on who assembles it, and on how many plugins they were talked into along the way.

And speed is revenue, not vanity. When Rakuten 24 optimized its Core Web Vitals and A/B-tested the result — same page, the optimization being the only difference — conversion rose 33.13% and revenue per visitor rose 53.37% (). Nobody redesigned anything. The page just stopped making people wait.

Speed is architecture, not optimization sweat

Look at the fast rows in the table. None of them are fast because somebody ran an optimization plugin the night before we tested. They’re fast because of decisions made before the content existed: how much JavaScript ships by default, whether the HTML comes off an edge node or limps out of a distant origin, whether each new feature arrives as another script tag with opinions.

That’s the bet StoryPress makes: pages served from Cloudflare’s edge, no plugin stack to accrete, SEO and consent and analytics wired in at the platform layer instead of bolted on at 40 KB apiece. Two of our three sites show exactly what that buys on a bad connection.

The third shows the limit, and we’d rather say it than hope you don’t scroll back up to the table: architecture gives you a fast floor, not a ceiling. Load 114 requests of enthusiasm onto a good foundation and you can still furnish a 5.2-second LCP. Our homepage is going on a diet, and we’ll re-run this exact script when it’s done — the methodology is published above, so you’ll know if we start grading ourselves gently.

Atomic facts

• Lab test, 2026-07-12: ten real small-business sites, throttled mobile Chromium (~1.6 Mbps down, 150 ms RTT, 4× CPU), median of three runs per site; one site dropped for DNS failure.

• Fastest LCP: crazyaboutcookies.com (Wix), 444 ms — also the lightest page (307 KB) and fewest requests (9).

• Slowest LCP: saltwaterbakeshop.com (Squarespace), 12.8 s on a 4.3 MB homepage.

• Best TTFB: Squarespace at 66 ms; StoryPress TTFBs were mid-pack at 248–326 ms.

• StoryPress LCPs: 664 ms (auresources.io), 1.5 s (make3.co), and 5.2 s (storypress.app — our own homepage, printed anyway).

• Worst CLS in the test: 0.91, on a StoryPress site (auresources.io); Google rates anything above 0.25 as “poor.”

• WordPress sites: 3.6–3.8 s LCP, 4.7–6.1 MB payloads, request counts in the thousands driven by ad tech.

• Page weight predicted LCP rank better than platform choice did.

• Field context: 74.86% of Wix origins and 70.39% of Squarespace origins pass Core Web Vitals, versus 46.28% for WordPress (CrUX Core Web Vitals Technology Report, November 2025).

• Business stakes: Rakuten 24 measured +33.13% conversion and +53.37% revenue per visitor after Core Web Vitals optimization (Google case study).

If you’d rather start on a fast floor than sweat your way up to one, StoryPress is $5 a month and built on the architecture that produced the good rows above — . For why speed increasingly decides who even gets seen, read ; for what the slow stack actually costs in dollars, see .