I run https://browserbench.org/Speedometer2.1/ about once per week on a 4 cores, Intel N100 box. It's not very fast, but it gives stable results.

I observed a steep regression on that benchmark:

Commit Score
1d9204b 52
578c52f 49
5580704 35

So we already had a small regression that I did not bisect between 1d9204b and 578c52f, but the changes in 5580704 are making things much worse.

I think we should revert that patch and figure out what caused the slow down.

cc @stevennovaryo