The pain point in this particular case (I worked with qhoxie and bjclark on this) wasn't so much that the EBS performance was consistently bad, but that it was

The pain point in this particular case (I worked with qhoxie and bjclark on this) wasn't so much that the EBS performance was consistently bad, but that it was very unpredictable. Servers would suddenly, for no apparent reason, start performing very poorly. That's been a consistent theme I've seen with EC2 instances--it's hard to predict how fast something is going to be, and once it's running, you don't know if its performance is going to change. That's one reason to take benchmarks of EC2 (and possibly other cloud providers) with a grain of salt. I've heard anecdotes of people starting up n EC2 instances, running benchmarks on each, then killing all but the fastest ones.