Well, my results are preliminary but I've finally run the comparison shots with the three cameras at 800, 1600, 3200, and 6400 (the 5D doesn't have 6400 so I underexposed 3200 by 1 stop and pulled it up in post-processing).
I've only compared the RAW files in DPP so far and here are my observations from comparing the ISO 3200 shots.
I took all shots at the same ISO, shutter speed, and aperture settings. I used the 70-200 lens, tripod mounted with the 2-second timer and mirror lockup. On the 7D, I shot at 70 mm, and on the two 5's, I had the focal length set at 110 mm. Not quite a perfect 1.6X, but very close.
I had to move the lighting back a bit with the 5D because it is more sensitive than the other two. At ISO 800 and f/8, the 7D and 5D2 needed 1/10 sec shutter speed, while the 5D metered at 1/13 sec. Instead of using different shutter speeds, I moved my lights back 3 feet or so, being careful to keep their direction the same. It would not have been fair to overexpose the 5D's shots by 1/3 of a stop as this would have given it an advantage.
Once the shots were taken, I loaded all the results into the computer and opened the ISO 3200 shot from each of the three. They appear to be equally exposed to the eye, and to the histogram so my lighting adjustment was correct. Next, I made sure that all my settings (sharpness, picture style, etc.) were the same and that I had shot all 3 with "incandescent" white balance.
I carefully compared all the shots at 100% in their native noise reduction states - none on the 5D (doesn't have NR for high-ISO), and Low on both the 7D and 5D2 (this is how I use them). I noted the following default settings on the DPP adjustment sliders from each camera at 3200:
5D Lum - 2; Chrom - 0
5D2 Lum - 5; Chrom - 6
7D Lum - 4; Chrom - 11
At these settings, the 5D2 is the cleanest followed by the 7D and then the 5D. The 5D2 also exhibits considerably more micro-detail even in its native resolution ( I have not resized anything yet). The 7D looks softer than the 5D2, and slightly softer than the 5D (perhaps a little tiny bit less contrasty). But, very small text is easier to read on the 7D due to the larger view on the screen ( higher resolution) compared to the 5D. When I resize them to equal pixel dimensions, I'm certain that the 7D will actually retain more microdetail.
Next, I made some adjustments to the NR settings in DPP. The first thing I noticed is that Canon's choices for NR in the "Low" setting on the 5D2 and 7D are pretty appropriate for each. I also noticed that the 5D reacts to rather low settings on the NR adjustment sliders in DPP.
I adjusted the settings in DPP to get visually equivalent amounts of noise in the shadow areas. These are the settings I came up with (there is a margin of error here):
5D - Lum - 2; Chrom - 3
5D2 - Lum - 4; Chrom - 5
7D - Lum - 4; Chrom - 9
At these settings, the detail presented by the 5D2 was still superior to both the 7D and 5D. The 7D still looked a little soft compared to the 5D2 but held more detail than the 5D by virtue of it's greater resolution.
My third test was to turn both the Luminance and Chrominance NR sliders to 0 on all 3 cameras. At this setting, the visual shadow noise was lowest in the 5D2, but the 5D seemed to be slightly cleaner than the 7D in the shadows. The 5D2 held its detail very well, and both the 5D and 7D were less detailed as before. The 5D held microdetail well and the 7D was slightly softer as before, but it's resolution advantage again made the fine print easier to read on the 7D.
My next test on this will be to return all 3 cameras to the NR setting selected by Canon when in-camera NR is set to "off" (these are not 0 on the 5D2 and 7D). I will convert them to JPG in DPP and then down-rez the 5D2 and 7D to match the pixel dimensions of the 5D. I should be able to crop the results and post them.
I have to say that it's pretty difficult to choose which NR settings to use before converting to JPG. It might seem fair to set all 3 cameras to "0", but that would remove the defaults that Canon puts on the 7D and 5D2. Plus, I really don't know if the 5D has some other type of in-camera high-ISO noise reduction that doesn't translate into DPP's slider settings. I've also considered setting all 3 to what appears to be visually best but I'm not sure that my choices would be best either. I could use ACR and convert them in PSE-7 instead but I like to use DPP since it matches what Canon has chosen for NR (for better or worse).