I wonder how much of Google's reliability comes from service failover and redundancy and how much from very reliable datacenters etc. I'd find it hard to believe that their platform and DCs aren't better than what we have seen from AWS, which could make compute engine a very attractive product.
A lot of it comes from automatic failover. Individual data centers have issues frequently, but if you're using High Replication Datastore then you won't notice it much, apart from occasionally seeing all your instances getting killed and restarted in a new datacenter, which also results in memcache getting reset.
