I've been recently experimenting with the idea of creating an open source S3 rest api clone that runs as a GAE application. This would allow for "private S3 instances" that leverage the google hosted infrastructure (big-table etc) and your existing S3 tools/applications/processes.
The project has reached a minimum level of functionality as a proof of concept, and is now hosted out on google code.
The core bucket and object-level operations are implemented, as well as support for multiple custom user accounts, ACLs and public-reads. More info on the site on how to configure third-party clients (smoke-tested with s3sync, boto, jets3t/cockpit, resourceful/spaceblock).
There are some obvious limitations right now due to the preview release quotas on GAE - e.g. object size limited to 1MB, and total storage limited to 500MB. The assumption is that these will be lifted or relaxed once the service is pay-as-you-go.
Would love to get people involved in testing and making the implementation more robust. I'm not necessarily a pythonista, so I attempted to write as little of it as possible...