We actually have tools to do all of that -- high-availability / fault tolerance / load balancing, entirely transparent to users, admins, and apps -- which we use to run Sandstorm Oasis, our paid managed hosting service. We are working on productizing these tools to sell to "enterprise" customers.

For individual users and small companies, single-machine Sandstorm is pretty reliable, but it's true we don't automate kernel updates yet (which will require scheduling reboots for off-hours). We'll add that eventually.

(Sorry for all the vaporware.)