Comment by rharter

timothyb89 is quite right. The cost wasn't just for a standard server, much of that came from the number of datastore reads I was performing, since the original backend wasn't using memcache the best, etc. It also wasn't syncing in the most efficient way, so that added a lot of overhead. There is still room for improvement, but just with a weekend of optimization, I've been able to reduce the cost by an order of magnitude. That, mixed with the automatic scaling makes it quite competitive.