Considering how often EC2 outages are EBS related, we've moved all our servers off of EBS to ephemeral drives. I'm surprised there aren't more people advocating this route.
I think part of the problem is that EBS issues also cause ELB problems, from that I read here on HN. I wouldn't know because we only use us-east for Hadoop.
On the other hand, our Cassandra cluster runs on ephemeral drives and it's way better than EBS even with the guaranteed IOPS thing. Everyone should definitely give this option a try.
