Or you could just run in multiple regions. Using multiple cloud providers limits your ability to take advantage of provider-specific features - why waste time writing your own load balancer when you could use ELB + multiple regions?
Not when the original goal of the very service is to have presence in all geographical regions. If aws us-east is hit, I want the users to transparently failover to a server on east coast (perhaps one hosted by google or softlayer) rather than be directed all the way to us-west or eu.
And as for ELB, one doesn't use ELB for a custom protocol that load-balances/fails-over itself from the client :-)
