One major reason not to hop on the cloud bandwagon just yet is issues like these. The more layers underneath that are not under your control the more layers you'll have to add to remedy that.

Systems with excessive complexity are hard to debug, especially when it comes to analyzing performance issues.

Given complete control of the hardware from the ground up it can already be quite hard to accurately pinpoint a bottle neck so you can solve it. Adding a lot of stuff between your code and the hardware is not going to make that any easier.

Typically a stack has 6 layers before you get to your application: drive, controller, driver, filesystem, database, app.

In a cloud environment anything under the filesystem layer is effectively out of your control and out of your ability to troubleshoot. The solution, to add another layer of complexity in order to combat the slowdown is really the opposite of what an ideal cloud environment would give you.

After all, the #1 selling point of the cloud is scalability and performance.

I think that it would be best if Amazon worked together with the OP to resolve the issue as a problem ticket rather than to try to solve it by adding a software raid.

Of course, that's just armchair reasoning, not being in the hot seat makes life easier.