During peak hours, the GPU pools are under-resourced and cause long pending times. This significantly increases CQ durations, since pending time has a triple effect on builds ['with patch', 'without patch', 'retry with patch']. 

For more detailed investigation, see https://bugs.chromium.org/p/chromium/issues/detail?id=913064#c6.

For related macOS issue, see https://bugs.chromium.org/p/chromium/issues/detail?id=898684#c3.

I'm marking the bug as Infra>Client since CCI likely needs to figure out exactly which pools are underresourced and how much we need to increase capacity. Please update if this should be directed towards foundation/labs.