The engineer in me can't help but look under the bonnet to understand how things work. While I agree exposure is a factor I have never found it to be a critical factor. There was a interesting series of articles on
www.dpreview.com last year ( which sadly I can't locate) which looked at the differences in ISO settings. For example the difference between 100 ISO and 200 ISO could be that a different curve is applied as opposed to cranking up the sensor gain. It is really a question of knowing the camera's native ISO. It maybe better to operate at 200 ISO as there is no real noise benefit at a lower setting.
Just as a crude experiment I took some test shots to examine the effect of under exposure. They were shot under room lighting at a constant shutter speed but different f stops @ 200 ISO on a Canon 40D. Similar sensor spec to Craig's 400D. each shot was processed in Photoshop CS3 RAW with noise reduction turned off and default sharpening of 25. The top row was correctly exposed, the middle row was nearly 2 stops underexposed and the bottom row was 3 stops under exposed.
One thing I noticed when I adjusted the exposure slider was a marked increase in noise level. However, I also noted that the contrast of the grey scale wedge also increased dramatically when the exposure slider was increased.
I suppose Photoshop does this because it assumes that under exposure also means the available light will give low contrast.
After adjusting the exposure slider I reduced the contrast on the 2 under exposed shots to match the correctly exposed wedge. Attached is a composite of the 3 shots. If you compare the top row with the bottom row there is an increase in noise but not that dramatic considering it was 3 stops under exposed.