I have asked this question here and on other forums and have not received many replies. I believe that this is simply because it is not known at this point. paulkk's anecdotal report is as good as it gets and notes from real users like him are very useful.
I was particularly interested to know the answer for the simple reason that I tend to buy 2nd hand. Testing a D-SLR body for bad pixels becomes one more thing to watch out for. It would upset me to find out that I paid $xx.xx for a body that had 200 bad pixels swapped out. Now, 200 pixels out of 6 or 8 million may not render the camera unusable, but it might affect the price I would be willing to pay. Since I am the one buying, I want to know. Checking for bad/hot pixels at the store is doable, but is there any way to determine if bad pixels have been swapped out?
I believe paulbk's argument that the usual post-processing will mask the problem, if the number of bad pixels is low. In previous threads, I have read comments from many users who are very easy-going about this issue. Some even said that they didn't mind if a brand new camera had bad pixels. That is farther than I would go, without getting a discounted price that is. If I paid $2000 (Cdn) for a new 20D, I'd expect it to work perfectly.
For all I know, new bodies have bad pixels swapped out at the factory as a matter of course. How would we know? What does their quality control consider acceptable? Zero bad pixels?, 10 bad pixels?, 10%?... I have never read an official statement by any manufacturer, though they may be out there and I simply missed them.
From what I can tell from the limited data, pixels do tend to go bad over time. Unfortunately this has a direct impact on people like me who buy 2nd hand equipment. As an amateur my cost/benefit analysis is completely different from that of a working pro for whom buying new every couple of years makes perfect sense. In general, they probably don't care about the issue as much.