All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=828928

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=ed3a27ad655f812ca0c902dc32148aff9cc0e4003afe6770c9c90cbf5e243222


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

pixel_xl

I'd expect to see a decrease actually, because the predictor *doesn't* write to the sqlite database during the period of the benchmark running, the data stays in a separate cache in memory. 
Anyway, 4K is just one memory page, I don't think it's worth spending much time investigating. I'm marking this as WontFix, please reopen if you find this issue important.

By the way, how did you determine that this commit caused the regression? I don't see bisect results here.

We use a local bisect script located at https://cs.chromium.org/chromium/src/chrome/test/vr/auto_bisect.py?q=auto_bisect.py&sq=package:chromium&dr since the VR Telemetry tests don't run in the Telemetry lab currently for a couple of reasons, which currently prevents the use of the dashboard's automatic bisect.