Bisect job to make sure it was V8 that regressed: https://chromeperf.appspot.com/buildbucket_job_status/9017951791928181872

Manually bisected it down to: https://chromium.googlesource.com/chromium/src/+/2e8266d16c0a6f6215e07761dec5ea937e637fec

Why did a Catapult roll regresses our benchmarks by 10 %?

Benchmark in question: https://chromeperf.appspot.com/report?sid=af540bf9d365af1e976546df0b8efd1f387f986ca264567c6baa5272db4d5915 (happening on all of the subtests.

ah, nice, that's the effect from the change that forces every benchmark to start on a cold page, so the regression is really the right base line we want to have.

+Zhen: nice work