Personally I'd like to see the photosite count go up to 64 million. Then I could, via RAW converter (or even in-camera), generate 16mp files using each RGBG Bayer matrix to create one output pixel. 16mp gives me enough spatial detail...I want finer, richer tonal detail.
-Dave-
The question is, if you are going to pay for a 64 MP sensor and the supporting electronics to digitize and (partially) process that data.... Would you not rather store the data with full precision to your memory card, leaving the possibility open to maximize the fidelity in a current (or future) raw converter?
I am not confident what "tonal detail" really is, but I don't think that in-camera downsampling is the way to get it. Whatever the camera can do (automatically and fixed) to your raw digital data, a software raw developer can do (with better precision, flexibility and user-interaction).
In my view, the "megapixel wars" is moving us further and further into the oversampled region. If the camera manufacturers can increase the number of sensels for a given sensor area without increasing image-level noise (they seem to be able to do the opposite), while lens manufacturers cannot improve at the same pace, then we will have more and more (accurate) samples of a 2-d waveform that gets (relatively) blurrier and blurrier (thus easier to sample and recreate). To me, that is happy-land.
-h