If you commit a large binary file and iterate on it a bunch you'll notice that your repo size explodes. That's why there have been multiple efforts (e.g. git-annex, Git Large File Storage) to improve performance with large binary files (by essentially taking the files out of the repo).
Big companies have issues with Git... here is an old thread that talks about what happened with Facebook and Git:
http://thread.gmane.org/gmane.comp.version-control.git/18977...
So they are patching Mercurial, apparently with some Google collaboration (!)
https://code.facebook.com/posts/218678814984400/scaling-merc...
git LFS comes to mind. https://git-lfs.github.com/
I've never heard that (and certainly don't personally deal w/ git repos that large). I thought massive repository handling was supposed to be a strength of git. How does git (start) failing at that scale ?