De-duplicating backup tool on a block basis?

view story

http://serverfault.com – I am looking for an (ideally free as in speech or beer) backup tool for Unix-like OS which can store deduplicated backups, i.e. only nonredundant content takes up additional space. I already looked at dirvish (my first candidate) and rsnapshot which use hardlinks to achieve deduplication on a per-file level. However, as I want to back up large files (Thunderbird mailboxes >3GB, VMware images >10GB), such file are stored again entirely even if just a few bytes change. Then there are rsync-based tools like rdiff-backup which only store deltas and a current mirror. However, as the deltas are gen (HowTos)