I think they were talking about just a small opt-in % of Youtube traffic, and that that percentage had a higher average latency than the rest of Youtube (vs shifting Youtube's entire traffic average).