Ah I see, so it basically enables a parallel run of webrender to gather compatibility/correctness stats.

So it renders it two times and compares the results?

It renders it in series -- stuff goes through webrender first, and then whatever webrender couldn't handle gets dealt with with the regular renderer.