Chirp suggests that Pver4 experiment on Win/CrOS stable uses about +1.5MB (of ~60MB) memory and +170 ms (of ~4800 ms) startup time. Figure out how to reduce this.  We could use the pver3 finch exp to measure any fixes, or we could measure it separately (with bots or with a different finch config after rollout).


These are a little weird since Finch has notified us of this same diff for a number of weeks, but then it shows up in "grey background" which means it's "NOT statistically significant."  So I'm not sure how real this is.

UMA metrics:
 Memory.Browser.Large2
  e.g. https://uma.googleplex.com/variations?sid=e92e0d506632d9e65e5c02b8c39883fb
 Startup.FirstWebContents.NonEmptyPaint2 
  e.g. https://uma.googleplex.com/variations?sid=96b657189846e127503bb4f920d87b81