One of the things that people tend to underestimate is datacenter outages. When an Amazon datacenter goes down it takes out many sites with it. We're working toward a world where data is redundantly kept in many different datacenters/locations. In those almost-total failure cases, latency is better than no availability.