bup is a backup system that makes clever use of a Git object store to create very efficient compressed and deduplicated backups. For a single arbitrary data point from a server I just backed up:

Plain filesystem: 13 G
Single gzipped tar file: 9.3 G
bup repo: 6.7 G

I restored all the files just to make sure everything was still there. (It was!)

bup also features one of the most entertaining and informative design documents I've ever read.

https://bup.github.io/
#bup

bup, it backs things up!

I continue to find that #bup is a great compression tool for large-scale data: I have an old image of a Linux root disk that has a lot of blank (zeroed) space on it. "Compressing" it into a bup repo was faster than gzipping it, and the result takes up less space than the .gz file. (xz with maximum compression gives a file 10% smaller than the bup repo, but takes much longer.)