[cratermoon]

It's not paying double. Take the total number of instances serving the app and divide them across data centers and availability zones within datacenters. This might require adding a few of servers if the app is replicated across a very small number of instances, but for apps that already have a significant number of instances to handle the capacity, it's just a matter of rearranging things.

Depending on how quickly new instances of the app can be brought up to, it may be worthwhile to run a bit more capacity so that the remaining instances don't melt down should all the instances in one zone become unavailable.