That would make the downtime less correlated, but if it was higher in aggregate, would it be a net benefit?

I'm talking about a traditional sort of model, where each project hosts its own issue tracker, code review, etc. Unless you're working across multiple projects, you'd only care about downtime for your own server, not other people's (except perhaps for dependencies, which should be cached locally anyway).

So where's the 'big point of failure'?

Unless you're working across multiple projects, it doesn't matter to you that a GitHub outage takes out everyone else's projects at the same time.

Can you beat GitHub's uptime with a self-hosted solution? Sure. But you'll waste more time setting it up & managing it than you'd lose from GitHub downtimes.