Speed and accessibility · Oct 13-14, 2026
Testing product-page speed without fooling myself
Lab speed scores swing wildly, so I tested each change against a rule set in advance, kept what held up, reverted what did not, and followed the evidence to the real bottleneck.
11changes tested, each against a rule set before testing
3–5lab runs per state; medians reported, never single runs
45 → 41active plugins after pruning
Method
Google's mobile lab test scored the same unchanged product page anywhere from 22 to 54, with main-image load times from 7 to 18 seconds. A single run proves nothing, so every change was tested the same way: a fresh browser tab per run, runs spaced a few minutes apart so the tool does not return a cached report, medians of three to five runs, and a pass or fail rule written before testing (for example, main-image load may not get more than 25 percent worse). Anything that failed was switched back and recorded.
Changes and decisions
| Change | Evidence | Decision |
|---|---|---|
| PayPal Pay Later and mini-cart scripts removed from product pages (checkout keeps them) | Median mobile score 32 to 40; main-image load 14.7 to 13.7 s | Kept |
| Skip a concierge widget's CSS and JS on product pages | Median score 35 to 45; load 13.1 to 11.8 s; page 71 KB smaller | Kept |
| Unload duplicate gallery plugin assets | Median score 28 to 31; load 14.9 to 12.6 s; one script error removed | Kept |
| Variations over AJAX above 12 options | Page HTML 1,009 KB to 508 KB; layout shift 0.154 to 0.033 in the second test | Kept |
| Text contrast CSS | Automated checker: 21 contrast failures per page to 0. Two earlier versions failed the speed rule; the combined version was kept, with one side effect written down (the disabled add-to-cart button no longer looks greyed) | Kept |
| One main landmark | Landmark errors 3 to 0 on two product pages; no measurable speed difference | Kept |
| Contact-form assets unloaded on product pages | No measurable change | Neutral |
| Image zoom switched off | Failed the speed rule | Reverted |
| Usage-tracking script switched off | Script file still loaded, no gain | Reverted |
| Gallery slider loop switched off | Worse score and load time | Reverted |
| Layout-shift CSS for the summary column | Score down 12 points | Reverted |
Every change was a snippet or setting that could be switched off in one click, and the list of what to undo was written next to it. Four plugins that duplicated work were deactivated, not deleted.
What the testing found: the server, not the plugins
- Six small plugins were tested with 20 paired timing samples each. None moved boot time by a significant amount, so deactivating them reduced risk but was not the cause.
- Any page that needed PHP took 1.3 to 2.2 seconds even when it did nothing, with spikes to 5 to 10 seconds. A cached page took 0.16 seconds and a static file 0.09.
- Site Health explained why: the opcode cache was full, PHP was an end-of-life version and there was no persistent object cache.
I wrote a brief for the host with the specific settings to change, and later moved the store to a rebuilt server running PHP 8.3, which is the largest speed lever and carries no storefront risk.
Tools
PageSpeed Insights (mobile lab), axe accessibility checks, WPCode snippets, WP-Optimize, browser dev tools and Site Health.
Part of my 2026 revisions to abundifind.com, a WooCommerce store I own and run. All revisions · Portfolio