We made the backend for the test open-source - it's a very straight forward Go binary that we run in several data centers around the world.
https://github.com/netlify/speedy
The test focus on global performance, which might not be relevant to everybody - but for ourself it really does! We have users all over the world and as we've been building out our product we've seen a pretty clear correlation between opening up a new PoP on our CDN, and starting getting clients from that region. The performance matters.
For the article itself - it's obviously not as lightweight as motherfuckingwebsite.com! That said, the 4.2MB download number is just plain wrong. We're currently transferring 1.2Mb on a page load when caching is disabled (everything is gzipped), but the critical path contents (the HTML + CSS bundle) is 68.5kb and once that's loaded the page starts rendering.
The lunr.js index is very nifty - click the looking glass in the corner to try out the instant search. It's of course cached between page loads.