Hi again,
FYI, here is exactly what we had in mind:
http://code.google.com/p/s3fs/wiki/FuseOverAmazon
It will transparently mirror files from S3 over HTTP on demand
So everything boils down to just have a bucket on amazon and mount it locally. IMHO, this is way more flexible and it can also be used by standard unix tools (do pre-fetch). That combined with the stats from EMS, you can have a very powerful sollution
Best regards,
Andrei