We (Sandstorm) had an HA/scaling story, which we called "Blackrock"[0], and it was (and still is) the basis for the Sandstorm Oasis hosting service. I actually think we spent far too much time on that part too early. It turns out we could have operated Oasis by running regular single-machine Sandstorm on a beefy (though not that beefy) VM and been fine -- honestly, we would have had fewer outages and better performance. We probably should have spent our time making the platform work better for developers and end users instead, and built out the scalability later when customers really demanded it...

[0] https://github.com/sandstorm-io/blackrock