You can't really create a massive co-located setup on demand for big jobs, then tear it down ... use the right tools for whatever job you're doing. EC2 being more expensive for your Postgres deploy doesn't mean it is an expensive toy
If you really have a need to setup and teardown a bunch of cores for the occassional large batch, nobody's questioning that the cloud is an economical way to do that.
But I've never had to do that. Not in a way that would make the development time involved economical anyway. Wait two hours for this once-in-a-blue-moon processing job to finish, or spend a day setting up a process to handle such jobs quickly in the future?
It'd really take something exceptional (again, with low IOPs demands) to have that make much sense unless I was already hosting in the cloud and had invested money in making such a task quick and cheap.
