Well, mainframe computers have such excellent uptimes (you almost never reboot one) because everything is hot-swappable. CPU failure? Remove the CPU module, insert new one, and continue - all while powered up. The OS takes care of suspending the failed one and scheduling around it. Ditto all other components. Effectively, you should never reboot them.
That's interesting. My recollection from working on them a bunch of years ago was that our mainframes were IPLed on a regular, scheduled basis, because the folks responsible for them were disciplined about it and wanted to make sure there would be no surprises when one needed to be restarted. The fault tolerance you mention was used to make sure that the scheduled IPL's were the only times they went down though, and I never saw any unplanned downtime on them.