Peter is right, the one way to make something fault tolerant is to design it to not fail. I recall many years ago in Paris, France I talked to a couple of developers talking about automating le Metro (as if a French labor union would ever allow that) and they were going to use 2 computers running parrallel and a 3rd to watch. I asked one question: "what if the watcher fails". Paralleling processors always present new problems i.e. you have an output from the board "what if the selector of which processor to supply the output fails?" The '51 exist in versions that will take almost anything (it was the code, not the '51 processor that made the Mars lander crash), look into those.

Happy hollidays,

Erik