I wonder if Glacier support in Duplicity will be possible without large changes. AFAIK, duplicity also reads some state from the remote end to determine what to backup (Although it also keeps a local cache of this?). To use glacier, the protocol would have to completely write-only.
I'd guess it would use a hybrid approach, with recent backups on S3 (which duplicity already does) being shifted to glacier after a period of time. The FAQ indicates that Amazon plans to make this easy.
