@LetTheRightLensIn: Perhaps we are not referring to the same sources of information and data. I can no longer tell, based on statements about low ISO DR having not improved from 5D II to 1D IV (which as far as I can tell from DXO data, it did. 5D II screen DR 11.16, 1D IV screen DR 11.46; 5D II print DR 11.86, 1D IV print DR 11.95. Given how bad low ISO read noise has been on Canon sensors, its not surprising they have trouble getting more than 12 stops.) First off, having re-read your previous posts, no...you did not say you were measuring SNR. Given that you are solely measuring DR, I'll certainly grant that is a much simpler test. But thats never really been my complaint.
Regarding the 5D II, I believe you can get more accurate results, because you can actually get your hands on one and create your own test images. Regarding the 5D III, unless I've missed something (posts regarding this topic are spread across several threads now), the best we have to go with are sample photos posted to the net (the best of which seem to be from IR at this point.) We can't be sure one way or another whether the darkest pixel of an IR sample shot is actually indicative of cap-on dark current read noise. At best, we can make assumptions. It seems you came to that same conclusion, which is why you used the black masked pixels in the CR2 file as a basis for minimum read level. I am pretty sure I have some bookmarks at home (and if I get the chance here at work, I'll see if I can search for them sooner) that have useful details about what those pixels are, what data they contain, and how to interpret that data. To put it simply, those pixels are specialty cases either for setting black point and/or calibration, and cannot reliably be used as a basis for minimum signal (there are some statistics about deviation around the base 1024 value they tend to have that you might be able to use to deduce a true dark frame minimum signal level...but even so, any conclusions based on such a measurement would not be particularly reliable as I'm not sure anyone fully understands those pixels.)
Given the lack of a sufficient set of test images, I don't think we (the internet community at large exploring DR, SNR, banding noise, etc.) can make enough conclusions to be concerned or relieved one way or another. I think doing so, and claiming that the tests and the results are as reliable as DXO's, is asking just a bit too much (and putting a certain responsibility on yourself that you probably don't want in the first place.) (Hence my rather rambling approach before to asking: Do you really believe you are being scientific in your measurements?) To put it more succinctly: Without known-good test images that we are certain contain patches of nothing but read noise in pixels other than the black masked pixel columns to the left and right of the sensor...we can't really come to reliable conclusions one way or another...about the 5D Mark III. Additionally, this is in contrast to conclusions about tests performed with the 5D Mark II, which we can be reasonably confident of given that we can actually get our hands on actual 5D II cameras in order to generate viable test images that contain the right data.
Apologies for being obscure before, I was not making my point clear. I'm not necessarily doubting the validity of your test method in its context (although I do think use of masked pixels is not reliable.) I'm doubting the validity of the test samples themselves as a means to calculate DR. As soon as someone gets their hands on an actual production (vs. pre-production) 5D III and can actually generate a cap-on dark frame and a cap-off channels blown frame, I'll be far more willing to listen to the conclusions as "reliable" and "similar to DXO".