A better method for safety critical systems is true redundancy:
Redundancy (engineering) - Wikipedia, the free encyclopedia[
^]
The hardware is duplicated, the software is written by different teams, and two out of three can "out-vote" the third. Purely duplicating the same calculation in different threads on the same processor doesn't provide you much security - it you think how many "soft errors" you get in the average year and compare that to the number of bugs you encounter in the same time... :laugh:
It's an idea, but I'm not convinced that the performance hit you would take would in any way balance the small advantage you would get.