This is a tricky one to pin down. The regression occurred during a period where our tests where broken, and the affected metrics changed within that period multiple times, not just for the visible regression.

https://chromium-review.googlesource.com/c/chromium/src/+/1208955 and https://chromium-review.googlesource.com/c/chromium/src/+/1233717 went in during that time, and the latter is what gets pointed to as the culprit. It's likely that whatever caused the regression went in between the original patch and the revert, making it difficult to bisect.