With the currently attached testcase, I noticed that Chrome gives large measurements on the first load (in the 1800ms range) and then smaller measurements thereafter (in the 200ms range) -- and they show the alert before the page has rendered.

So I'm suspicious that they might be firing onload before the first render, which might not make the current testcase an effective measurement of time-to-finish-the-layout.

Here's a modified version with a more explicit layout flush before the alert.

Sorry for not circling back on this for a while!

Good news, though... It seems to have improved since this was filed.

Using the attached "testcase 2", here's the range of measurements I'm seeing from various browser versions (performing 3 attempts with each):

Nightly from 2016-09-01: 5680-6000
Nightly from today: 1100-1350
Chrome: 700-1000

And using the original page ( https://lodash.com/docs/4.15.0 ), both Firefox and Chrome seem to react near-instantly when I change the documentation version in the dropdown now.

So I think this is largely WORKSFORME. Chrome is slightly faster, but not extremely perceptibly so, and not by an order of magnitude like before.

I'm closing out this bug since we've mostly closed the gap (per comment 8), and I filed bug 1523432 on the remaining small performance difference between us & chrome here.

Thank you!