How we reduced page load time by 4 seconds on our startup

This is contrary to received wisdom, and I love hearing contrary takes, but the graphs shown in this post do not suggest that it is true. The post then spends a whole lot of time explaining hard things to do which might theoretically improve this, and very little time explaining the easy things -- on-page factors like covered in the YSlow presentations -- which were apparently left uncompleted. That is a shame -- for the amount of work required to even start thinking about caching you can have spriting done, and that would have knocked off much, much more than the 25ms that was successfully knocked off the time to first byte.

You can test this pretty easily: add a 250 ms sleep to the page. I'm guessing that it does not add much to the user perceptible delay.