That's a very smart solution indeed. I wonder if it is also possible to throw more memory at the problem? Managing instances with locally attached disks on the cloud is a bit of a pain, as you don't have the capability to stop/start the instances anymore.
