Dev.to WebDev 🛠 Dev 👁 0 📖 3 min read

I tested my own WordPress plugin against two cache plugins and published the null result

I publish a small WordPress plugin on wordpress.org. Peakspire Font Loading Optimizer does one thing: it preloads locally hosted WOFF2 files, capped at four, so the first font file starts downloading earlier. Its readme

I publish a small WordPress plugin on wordpress.org. Peakspire Font Loading Optimizer does one thing: it preloads locally hosted WOFF2 files, capped at four, so the first font file starts downloading earlier. Its readme says it is safe to use alongside cache plugins. In November 2026, at version 1.2.0 and 400 downloads, I stopped trusting my own sentence and tested it.

The question

Preload tags live in the head. Cache plugins rewrite the head: they minify HTML, combine CSS, and serve stored copies of the page. Any of those steps could drop, mangle, or duplicate a preload tag. So the claim "cache plugin safe" is really six test conditions: my plugin on and off, crossed with no cache, W3 Total Cache, and WP Fastest Cache.

The test: those six conditions in a local lab. WordPress 6.9 on PHP 8.3, Apache, no other plugins. On one side W3 Total Cache 2.11.0 with page cache, HTML minification, CSS minify and combine all verified active. On the other WP Fastest Cache 1.5.5 with its cache, minify HTML, minify CSS, and combine CSS options verified active. Then 40 interleaved Lighthouse mobile runs across the cells, alternating conditions so a slow afternoon could not fake a win.

Getting WP Fastest Cache to actually turn on was its own lesson. Its master toggle checks for the literal string "on" from the settings form, and even correct form posts dropped the status key. The reliable path was a direct option write in the exact persisted shape. I only believed it was running when the cache files materialized and the page bytes changed.

What held

The headline finding is boring, which is the point. The preload tags survived every cell. Through W3 Total Cache's cached, minified, combined renders and through WP Fastest Cache's, the tags came out byte for byte intact, crossorigin attribute included. On a theme that already self-preloads, the plugin added nothing, which is the correct behavior: no duplication, no reordering.

The speed finding: the first font file finished about 620 ms earlier with the plugin on, roughly 27 percent. That is the whole performance claim, and it matches what the plugin is for.

What did not move

Largest Contentful Paint medians differed by less than 100 ms across all conditions, which I read as no clear improvement either way. I have no LCP win to sell. Preloading the font earlier helped the font fetch, but that earlier fetch did not translate into a clear LCP improvement in this test. Publishing a null result feels odd when the download graph is the only audience, but a self-audit that only reports its wins is marketing. This one reports both.

The test has limits and they are on the page: one theme, one lab host, free tiers of both optimizers, local network. I am not claiming your stack behaves the same. The method is the deliverable: name the claim, build the matrix, interleave the runs, publish the cells including the flat ones.

Why a 400-download plugin deserves this

Nobody was asking. That is exactly why it was worth doing. The WordPress ecosystem runs on readme promises, and the cheapest trust signal a small author can offer is a public test of their own promise with the negative finding left in. The full matrix, the run logs, and the limits are on the proof page: peakspire.ca/font-plugin-cache-compat-test/

If you ship a plugin with a compatibility claim, the test is an afternoon plus a lab container. The null result costs nothing but the afternoon.

📰 Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.