My take-away message from this: if you want predictable predictable performance using AWS you need to use largest instances and EBS block sizes. This suggests Amazon scales up but not down. As in, if you're using the smaller instances (because you're a startup, say, and don't have a great deal of traffic) you're hosed. On the other hand, a lot of the calculations I've seen suggest maintaining your own server infrastructure is cheaper than AWS and you get more predictable performance, so I wonder why anyone who has a lot of traffic would use AWS (except for handing traffic spikes or batch jobs). So to me it seems that AWS doesn't have a great use case. I'm interested in other viewpoints.