Finch says there is a significant regression:
https://uma.googleplex.com/p/chrome/variations/?sid=a4e5a70183793352086ab75a1ebb9a73

50th %ile ~6% regression^ which is ~2.5us
75th %ile ~6% regression^ which is ~10us
99th %ile ~13% regression which is ~700us
^ statistically significant

This regression is present in both canary and dev and across platforms.

We have some microbenchmarks of paint (blink_perf.paint / fixed-and-many-layers-scroll) but it may not cover what finch is showing. Tien-Ren addressed some of the microbenchmark regressions in  https://crbug.com/818768  and the finch data on canary in the past 3 days looks slightly better but it's not statistically significant yet.

Anyone have ideas? An option is to wait for more data at the end of the week and potentially merge all of Tien-Ren's patches (just 1 of 3 were merged to beta).