In the source:

   
    
      notice
      
javascript required to view this site
      why
      
measured improvement in server performance
      
awesome incremental search
    
  

Indeed, that was a relatively recent change that I never understood. It had been working without JS for a long time before that. And page loads became slower, not faster, for me.

> And page loads became slower, not faster, for me.

They're not exclusive, smaller server loads doesn't mean the experience is better for the client.

For instance the server might just stop doing server-side template rendering. Instead it sends one static page, then the client requests and render the JSON for the article, or maybe the page only contains the JSON for the article instead of the rendered one.

Then the template rendering cost stops showing up on the server, because it's not performed on the client, and serialising to JSON is almost certainly cheaper than rendering an ad-hoc and half-assed template format.

But now the client has to perform a bunch of extra requests (to fetch the javascript and possibly the json) and it pays the cost for the template rendering, and that might be even halfer-assed, and more expensive, because it's out of sight of the server loads / logs.

The result is that the server has gone from, say, 80ms CPU to 40ms CPU per page (which definitely qualifies as a "measured improvement in server performance"), but the client now needs an additional 300ms to reach loading completion.