It's actually not hard to implement if you separate the index file from the blobs... but the append part probably won't be feasible with S3