I am not sure why the metrics could report a larger number. If anything, it should be less as the time we get from fence should be a time in the past.
@klausw, do you have any suspicion? The fence already exist before my patch. And the regression is about 2ms.

Oh, I remember it wrong. The old report use heuristic to guess the time. And after my patch, we use the accurate time. The heuristic is something like this:
      // Fence was already complete when we checked. It completed sometime
      // between the saved webvr_time_copied_ and now. Use the midpoint of
      // that as an estimate.
      base::TimeDelta lower_limit =
          rendering_frame->time_copied - prev_js_submit;
      base::TimeDelta midpoint = (lower_limit + prev_render) / 2;
      webvr_render_time_.AddSample(midpoint);
So it is probably just because the previous heuristic was report the wrong number. If so, there isn't anything that we should do.

close this one. Klaus and Michael, if you disagree and think that there is something else, feel free to reopen.