You'll probably want to look into using CEPH/similar when/if the service starts to get busy. Partly for redundancy but also if all your repositories are on a single shared disk you're gonna hit I/O contention if many people are pushing/pulling at the same time - even more so if you get any CI system wired up.