All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=591757

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDA36OYvwoM


Bot(s) for this bug's original alert(s):

android-one

Looks like https://chromium.googlesource.com/chromium/src/+/f5ff205e2de07090a8dd52397b2573368fdc3229

Although https://bugs.chromium.org/p/chromium/issues/detail?id=588354#c6 states that there were fewer perf regression with this approach, this still looks pretty bad.

keishi@, is this regression expected?

Feel free to mark WontFix if you think this is reasonable.

Yes as I noted in the bug for the CL, page_cycler.basic_oopif has regressed pretty badly because of it. In the old code we were triggering GC for each iframe creation so my CL reduces that and a slight regression was expected. If you looked at the per page measurments, it looks like nationalgeographic is the only one that regressed. It looks like that particular site may have reacted badly with my change, but since it didn't regress other pages I think this is acceptable.