I think you read that backwards. In skydhash's test, Codeberg's data was 72% cached, and GitHub's data was 28% cached. Maybe you meant that GitHub's cached 4.28MB was, in absolute terms, more than Codeberg's cached 1.41MB?

Some parts of Github are SPA island, which is why the DOM load fast, but then it has to wait for the JavaScript files and the request made by those files. Codeberg can be used with JavaScript disabled and you don’t have that much extra requests (almost everything is rendered serverside).

The transferred part is for the gzipped transfer. That makes sense if the bulk of the data is HTML (I have not checked).

I’ve disabled the cache for the network requests.

Oh, thank you for the correction. That was a dumb mistake on my part.

Yeah, that is what I meant. It looks like Github's strategy is to push all the initial data they need to cache, to optimize subsequent requests.