I'd not hold my breath for it.
We evaluated that one very closely and the diffs we get from ZFS do not work for the sliding window algorithm in use (which is only semi-correlated to files), it would effectively mean that combination of deduplication and full-backups would need to be abandoned, which we do not want - backups where one needs to apply multiple diffs and pray that that works out is something we want to avoid at all cost.
Ceph rbd would also only help for fixed size block based backups, as there's no function which can map block diffs to a dynamic block sized archives. There was lots of technical discussion, on lists and also internally - if there's a break through we will also certainly go for it; but it currently does not seem like that would be anytime soon.
Aww that's too bad, thanks for the detailed explanation though.