I'm now doing an initial backup with dirvish, which is a far superior backup program to the one I used previously, rdiff-backup. rdiff-backup is basically a Python script that stores binary diffs between backup increments, so you have to wait approximately three billion years for it to unravel its own diffs in order to restore to anything but the most recent backup.
Dirvish uses filesystem snapshots instead (usually hard-linked trees, but I'm using a patch that allows it to make use of btrfs copy-on-write snapshots for better efficiency), which means that each backup increment just appears as a directory on my filesystem. I can browse each increment as I would any other filesystem tree, and even trivially grep across multiple historical snapshots if I'm looking for something specific.
The other really nice thing about dirvish is that it's very easy to specify custom snapshot expiry. Previously, I kept backups going back 90 days, but I think now that I have dirvish I'll keep some backups (say, one increment per month) forever. October 2014 will be the first month I always have backups of.