I just got my 5D Mark III last week on Thursday. Took me a while to shoot some preliminary tests.
I've always been that rare professional that preferred DPP to Lightroom, mostly because I've had an image database program running in my studio since 1996, so that a big chunk of Lightroom is superfluous for me and, as to the actual heart of the program, the raw converter, I've found DPP to be at least as good, if not better at producing good 16 bit Tiffs, my file of choice, at least the one I start out with. Although Lightroom has more refined tools than DPP, the basic conversion is not always as good and the program itself is 10 times biiger with what seems to be unnecessary bloatware included. So, when testing my new cam, I wanted to use DPP, not Lightroom. I wrote Bryan Carnathan at the Digital Picture to ask him his opinion of the new version, 18.104.22.168, and whether it "solved" his softness issues with the 5D3 output. He very kindly mailed me back, saying that, yes it did. So, I used my updated copy of DPP to convert my tests. I also, just because I wanted to see what it could do, used the Digital Lens Optimizer on each file, found under the Lens tab in the tools pallette. I'm not sure how much more this Optimizer does than what is now automatically done in-camera in the 5D3, but I used it anyway, and didn't yet bother to compare results with it and without it.
My conclusion is a subjective comparison of these raw conversions with what I am used to getting with my now gone- but-not-forgotten 1Ds Mark III, not the 5D2. From what I know, the low ISO results from the 1Ds3 were, if anything, better than those from the 5D2.
Here are my observations. The 5D3 focusing system is better than that of my 1Ds3. It nailed static subjects (didn't yet try action) more precisely and with the 2 L lenses I tested, (24-70 f/2.8 and 16-35 f/2.8 v2), with no lens calibration. My tests were shot at f/4.0 and f/7.1, depending on the depth of field I needed. The files showed a small but significant gain in low ISO (shot at 200 ISO) DR. The exposure meter in Eval mode semed to give me uniformly better exposures without needing to "outsmart" it quite so often. The shutter release actually felt a little better and the seemed very slightly better dampened - a very unexpected result. The images were mostly quite sharp looking, but since my exposures today seemed to yield very narrow but nicely shaped "waves" that filled the center of the graph space but did not extend all the way to the shadow or highlight sides on the horizontal axis, they seemed to lack some contrast; this was, rather than some fault of the camera, rather the result of my exposure choices, (mostly) the existing light values, the default tone curve in DPP and the slightly better DR of the camera. In Photoshop, when I applied some very minor curves and then very minor low amount-higher radius-zero threshold unsharp mask to the image, the histograms spread out as far to the left and right as was appropriate and the image popped like a bubble gum ballon, without any visible artifacts. Most, if not all, of this could also have been accomplished in DPP or Lightroom, if you didn't happen to have Photoshop. Overall, the resulting files looked at least as sharp as any result I could have had with my old 1Ds3, and they have just a little bit more detail in them as well. Finally, the defringing and other functions in the Digital Lens Optimizer seemed to work so well that I can honestly say that it brought "new life" and a whiole new look to my sometimes chromatically challenged 16-35 f/2.8. It couldn't turn it into a Nikon 14-24, but it made it look a heck of a lot better.
I'll not be posting these, but I think I may have given you all a pretty good description of my own personal subjective findings. They're not scientificly rigorous, but they are backed up by my experience of over 16 years spent photographing, scanning and retouching digital files for very very high end commercial clients.