Basically once the design is proven, then you can move away from certain tests. Just to give you an example of a more destructive test - drop testing....
When you have a new gun design, you can drop test it many times. It may break. You change the design and do more drop testing. Say it passes this time. You now know that the design is strong and it passes a drop test. You don't drop test every one you make after that. And true, you never know for sure if the next one would have broke or not.
I know this is not a perfect analogy because drop testing does physical damage, but my point is that, once every unit you test for a while does not break, then you should probably stop that test and spend time on things which are more likely to find bad parts.
I am claiming that this is not 50 year ago, and AR bolts are to the point where you can test thousands of them in a row and not have any failures, so one is better off spending the dollars for that test on additional dimensional testing by upping the number of statistical samples you take which will increase the probability of locating bad parts (dimensional tests commonly finds bad parts, so it is a much more productive thing to spend time on).
Why not do both? It could take 50 or 100 man hours to test everything there is to test on a rifle. Realistically, you can budget up to an hour or two at most, so you really have to spend the time on what finds the most bad parts. MPI does not find the most bad parts.

