Yes. In practice it can be worthwhile doing this in distributed compute. These methods (and those from sibling comments) do rely on the error rate being low, and the data compared being much smaller than the amount of compute (eg bytes in final answer << number of CPU instructions used). AFAICT such faults can be treated much like network and storage errors.
What we can’t do is have a set of CPUs where roughly every 1000th instruction fails and hope that plugging together a bunch of computers together to check each other’s progress will work. The specific issue is that the checking is wrong so frequently that you can’t “win” to solve a problem involving trillions of instructions by adding more computers. The overhead of “checking the checkers” just blows up.
What’s interesting about quantum computation is this is exactly what is proposed - have qubit error rates of (I think) around 0.1% and lots of measurement, classical compute, and feedback control to keep the computation “on track”. The whole scheme relies on the error rate for all of that classical compute step to be negligible.