long story short
nice that you have seen what I have tried to explain about read noise and banding since I started as a member here
Just to prove my point. Here is a full-sized JPEG of the "best fit" shot from my living room. The highlights are still blown...probably by about three stops (as this was one stop down from 0EC, and it took five stops down to fully recover the highlights). There is about a stop of super deep shadow detail that is also completely lost to noise and thus unrecoverable. There IS noise, even some banding, so we are definitely at the limits of our DR. Unlike the most extreme example before, which tried to capture the entirety of the full 18 stops of DR in a single shot and expose the worst qualities of a Canon sensor, this is a the best I could do without putting a Canon sensor in the worst possible light. This is a "reasonable" recovery. It is not as good as a D800...we wouldn't have lost the highlights with the D800, only a little bit of the shadows. But it is not the "worst case possible" scenario you/Mikael have frequently put forward:
Full size (100% crop): http://i.imgur.com/VAmOF7c.jpg
BTW, this image has no noise reduction of any kind. With a little bit of basic luma NR and some Topaz DeNoise 5, those shadows can be recovered quite nicely. We are still at least four stops short of fully resolving the DR of the scene, so noise reduction will only take us so far. We might gain a stop of DR with NR. The clipped highlights are gone for good, but we can probably gain on the shadow end. If one is willing to take NR to the limits, and spend the time to extract every last ounce of detail, you could probably get away with another stop, maybe two of underexposure and still recover. It won't be as clean as a D800, and will probably still have some clipped highlights, but it also won't be riddled with the nasty red banding that is the hallmark of Mikael/Ankorwatt's attempts to prove how terrible Canon cameras are.
Hopefully this rounds out the balanced argument.