This is a good idea: Amazon has integrated Hadoop with a S3 data store back end. I think that this will be great for companies that only need to occasionally perform large scale parallel data processing. Server instances are billed in one hour increments so with some experience, it may be possible to estimate how many server instances to rent to get large jobs done using just less than an hour.
I have been enjoying using Amazon EC2, first on a customer job, and more recently preparing a custom Amazon Machine Image (AMI) with all of the example programs and systems pre-installed and configured for my new book "Intelligent Scripting for Web 3.0" for APress. It looks like everything will run fine in a small server instance so readers of my book can experiment with the example programs for ten cents an hour. (Or, spend the time to install everything on their own development system.) I am working on Chapter 13, so just two more chapters to finish :-)