power.gpu_rasterization.top_10 occasionally fails on Facebook.com because it doesn't become network-quiescent in 60 seconds. I suspect this is some problem with the code on that site which keeps reloading resources indefinitely and not an issue with the benchmark or Chrome.

I'd rather not remove Facebook from the page set. Should we instead still perform the measurement to get a number out instead of making the bot red?

This benchmark doesn't appear to have an owner. Juan, maybe you have some system health perspective here?