The problem is that most people are not going to take the time and trouble to make prints from 8 bit and 16 bit processrd files to see if and when there's a noticeable benefit to using the extra resources required for 16 bit.
There is an assumption that, if you can't see it on the monitor at 200% then you won't see it in the print. In fact, I think most people's experience is that stuff you can see on the monitor doesn't always get translated to the print. I believe this is because a view of a detail on the monitor at just 100% often represents a print size much larger than what's printed.
When CS3 was released, I was at first undecided whether to pay the extra money for the CS3E version. After experimenting with the stacking options in CS3E, I decided it was worth the extra money and I set out on my recent trip a few months ago with a copy of CS3E on my laptop.
One experiment I did was to photograph my hotel room with my 5D without using flash. To get maximum DoF, I used f16 which at ISO 3200 allowed for a shutter speed of only 1/20th second. The lens was the Sigma 15-30 which doesn't have IS, so the image is understandably not particularly sharp. However, never mind! The purpose of the experiment was to see how CS3E's stacking options and auto-alignment feature would handle a series of hand-held shots taken in continuous mode, bearing in mind the slow frame rate of the 5D.
I took a burst of 7 shots. There's some slight movement between frames as one would expect, but CS3's auto-alignment seemed to handle that perfectly.
What I couldn't do, however, was stack those images in 16 bit mode. My laptop didn't have the resources. 2Gb of RAM is apparently not enough. I had to convert all 7 images to 8 bit tiffs before stacking.
Now that I'm back in Australia, and after reading this thread, I thought it might be interesting to stack those same images again in 16 bit mode using my desktop computer, and compare.
Below are 100% crops comparing, from left to right, 8-bits stacked, 16-bits stacked and a single unstacked image showing the noise difference, followed by the full scene.