This Nikon one seems to have a poor ISO which is odd.That would be odd; what are you basing this off?
For those of you worried about Canon's response I would like to point out something. TheYour major assumption seems to be that it is no more difficult or expensive to create a sensor with a larger surface area than another, but with the same pixel pitch (and also that the size of camera pixels isn't dependent on the production process, and also that there is no improvement in production processes in two years). Even with two years' development time, this could (I can't say for certain, not knowing the ins and outs of Canon's tech roadmap and which facilities they use for which processes) require using a more expensive process than they currently use for full frame sensors (of the same generation as the 7D, that is) and they also probably would have to lock down the reject rate (I'm not saying it is cheaper to have lots of rejects if they only take up a smaller part of the silicon wafer at the plant, however).
7D has a 22.3mm x 14.9mm sensor (area of 332.27 sq mm) and takes images that are 5184 x 3456 (17915904 pixels).
17915904 / 332.27 = 53919.72 pixels / square mm
A full-frame sensor is 36mm x 24mm (864sq mm). If Canon used the 7D sensor technology that packs 53919.72 pixels/sq mm on a full-frame sensor it would be 53919.72 * 864 = 46586638.08, which is a little over 46.5 mp.
Canon, using 2 year old sensor technology, could release a 46.5mp full-frame camera with 7D ISO performance. I wouldn't speculate on whether they will choose to do so.
UPDATE: Another thought occurred to me. 46.5 is a little more than twice the 5D Mk II. Wouldn't it be interesting if there was an HDR mode to this sensor which took a 23.2mp image with half the photosites set to one ISO and the other half at a higher ISO?
By the time all that work has been put in, what I think they're actually going to do is just produce the sensor with the best qualities that makes use of the current production technology, which will give them the smallest feature size at no higher cost. It might not cost much to keep older tech lines going a while longer, but if the 18mp sensors use an older, "coarser" production process it might actually cost them money while not putting them at the forefront of the race with technology.
To pull all the technobabble out of the post - if Canon now uses newer factory machinery to make their sensors than were available for the 18mp sensor in 2009, the features of the new sensor will be mostly determined by that new process. They would not use the older process because it puts them behind in technology while giving them no real benefit in cost savings because everybody else is rolling out newer tech as well.
All that junk aside, I think you're onto something - as a baseline that is probably easily achievable for Canon given it already has had something out for a couple years with a pixel pitch that would let you make a high resolution sensor, as you said.
As for your HDR mode idea - that's an interesting idea, but I don't think it would work with current technology, unless you want your pictures to have the appearance of an interlaced or dithered image (at best). I think the DIGIC's probably equipped to process pixels the fastest when all pixels are assumed to be at the same ISO - of course this doesn't mean it isn't possible but it may end up being very slow or even cause some accuracy problems, as before the DIGIC can work with the raw photosites they have to undergo an ADC process, and again I would wonder if the design is such that you can just specify some exact, arbitrary pixel is going to have different gain applied to it than the last and next ones.
The bigger problem is how to mesh what are actually different images taken from different locations into one, because in your model you have to offset one image to "fit" the other, which means interpolation, which further means inaccuracy and real data is lost. In your theoretical HDR camera, the "second" frame of DR is either every other pixel, which means it's not taken from the same location, and different details will be recorded (as I said - this would be worse than halving resolution), or it is every other line, which is funky as well.
All the same I would love it if they did some pixel binning or something similar to help extend DR, if possible.