For a reasonably common class of deployment and organization (E.g. an online service where the cost of a short outage is significant) it becomes a pain point. For instance if there's a sudden security flap and you need to re-spin your system and redeploy, but that process is gated on GitHub working, now you're screwed if that security flap happens when GitHub is in its 5% down time. Basically you've coupled your uptime to GitHub's uptime in some measure.