Derr so you can be sure the output is correct? You don't run tests in a live environment when you're trying to quality check a product... that's QA 101. You run your tests against a known good data set with known results and compare what you get from the unit.
I totally agree with you two, but you don't need to run that for days on end it's a machine once you see once that it's giving you the results you want, you know it won't give you bad results unless the device fails for some other reason after. After a run on a tester for quality control, running it on live environment would both continue to test the device and aliviate running costs, since device failure is higher shortly after production then after it's been in operation for a while and they cost less to run than what they make in the same period.
That's not necessarily true.
When someone is testing a new CPU overclock, what do they do? Well, they might run Prime95. But they don't just run Prime95 for 5 seconds, then call it good because "it is giving the correct results". No, they run it for 24 hours or longer, to be absolutely certain that the processor is stable for long periods of time. Some people report errors as far as 15 hours into the testing.
So, BFL's testing period is 12 hours, after which, the units are packed up and shipped out.
Now, if you want to argue that BFL is purposefully holding on to units in order to get an extra $2.50 a day, well, to me, that's just silly. But either way, it's purely speculation and goes directly against what BFL says they do.