We provide this long term for backups without excessive space usage by ignoring many classes of files that are expected to be very uninteresting, such as very large file, automatically created files (dot-o files) and other classes. This is not possible with snapshots which are very much an all or nothing approach.
Exactly what scheme we will use when this filesystem goes live is not yet clear, but I suspect a heuristically limited incremental backup scheme similar to our current scheme, but using the high-bandwidth interface that was created for archival backups will meet our needs quite effectively.