I'd be surprised if the answer wasn't latency - this kind of server load would be easy to scale with their engineering resources and they're big on the competitive advantage of user-perceived performance.