It happens that we get an unrealistic fast TTFB in our tests in our tests because of problem with SPDY in WebPageTest. It happens in first run and second page view but I think they are unrelated.
- We should verify that we get wrong TTFB for our first runs (it could have been fixed in Chrome 49, WPT uses the timing event log and there's been a fix there).
- If we still have the problem we should choose to pick the median run instead for a while, because the bug doesn't always happens so it is a bigger chance we get correct values if we pick the median run.