In my experience micro instances are not suitable for on-demand services since they have no reserved CPU and frequently stop responding for minutes at a time. They're basically a way for Amazon to sell excess RAM to grid computing customers. On the other hand, I tested them a few days after they first became available and was just experimenting with them as a way to test scalability of a product, so things may have changed since then.