If they're running this under sustained load for days, could that not shorten the lifespan of the machine?
Whether or not that is the case, the user still ends up with a refurb but without getting the refurb discount.
I think, for a newish machine (up to say 6 months or so), they should replace it without hesitation. These things should work right out of the box.
If they actually did stress test the machine for days, then I'd agree with you. My experience was that the diagnostic tests didn't replicate the problems because it was clear they didn't run them for that long. It was only when the display errors happened right in front of them while I was there that I got a GPU replacement. Subsequent to that, when the coil whine problem developed, it was having a recording of the noise that did the trick in getting them to replace the whole machine since they couldn't reproduce it themselves either.
The point is that Apple in fact does do the right thing once the problem is identified (replaced hardware, replaced machines), but the diagnostic processes at the Genius bar don't work well, particularly for the nMP. Also nMP users are too quick to blame software bugs for intermittent hardware issues and because they're using these professionally, there's a legitimate concern about being without one's machine for a week.