A CL in this revision range caused a ~2ms increase in WebVR frame render time: https://chromium.googlesource.com/chromium/src/+log/e9d46063e6c95c205b8fa45c1bd1d50507d20fc8%5E..c91d4ca6224b4284a6ae14c08efddbb1144de67d?pretty=fuller&n=1000

Somewhat strangely, this only affects a single test page: https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgIDgi4aUpQoM

Frame JavaScript time looks like it may have also been affected by this, but the test run immediately following the one where the regression showed up included the accidental fix for the slow/fast render pattern, which decreased frame JavaScript time significantly, so it's hard to tell.

Looking at the trace for the run that caught the regression and the one right before, the median values look very similar (~20ms spent on rendering before and after), but the later one has a much larger spike in the beginning. Previously, rendering time spiked to ~40ms at the beginning of the test, but the regression causes that to increase to ~340ms, which has a significant impact on the average. So, this looks like it's some regression in VR startup and not sustained performance.