Startup.FirstWebContents.NonEmptyPaint2 has a regression with OOP-D enabled on mac, primarily near the 75th percentile.

https://uma.googleplex.com/p/chrome/variations/?sid=a9314e3430429ca111dbe1fb5f4d1e64

Startup.FirstWebContents.NonEmptyPaint2 actually has a 1-2% regression on Windows (and presumably all platforms) that has to do with how it's measured, which I started to look at in https://crbug.com/895588. The regression is larger on mac though so there might be something else going on. At 75th percentile OOP-D is adding an extra 2 seconds before the first non-empty paint.

Startup.BrowserMessageLoopStart.To.NonEmptyPaint2 is submetric that measures the time from browser main loop start to Startup.FirstWebContents.NonEmptyPaint2. That has the same regression

https://uma.googleplex.com/p/chrome/variations/?sid=75698fbc9f3a27367cda1100b912e9f4

Looking at timeline on dev channel OOP-D has been consistently worse here:

https://uma.googleplex.com/p/chrome/timeline_v2/?sid=264905012e4f732546eced501cd2305a

This is probably worth investigating further to see if we can understand the regression better and maybe improve things for M71/M72.