Some remarks on the Noise/DR/Resolution section of the article:
Resolution vs pixel level noise is always the tradeoff, if one's criterion is noise at the pixel level. That is simply because noise power rises with increasing fineness of scale in the image. All other things being equal, decreasing the pixel size to increase resolution simply samples the scene at finer scales, and noise at those finer scales is necessarily higher due to the physics. For instance, here's a plot of noise power vs image scale (spatial frequency) for a 40D and 50D (test images from Imaging-Resource; ISO 1600, converted from RAW in DPP):
The horizontal axis is scale in the image, the vertical axis is noise power; so each data point is a measure of the amount of noise at a particular scale. The somewhat arbitrary units for the horizontal axis put the Nyquist frequency (the limit of resolution) at 256 for the 50D, and at 209 for the 40D. The two cameras are more or less the same up to the point where the 40D stops resolving, while the 50D climbs a bit higher. So, the pixel-level noise in the 50D is higher than that of the 40D, simply because it resolves more. If one were to downsample the 50D image to the 40D pixel dimensions (I have done this and shown results in some threads over at DPR), the 50D plot would look like a clone of the 40D one -- the fine scale noise is thrown away together with the extra resolution upon downsampling. So the noise and the resolution come hand in hand -- one simply has to decide what one wants.
As for the noise being less "grid-like" and more "stochastic", that is a property of the way the RAW converter is interpolating the Bayer data, and has little to do with the capabilities of the camera. For some reason, Adobe products (ACR/LR) are extremely variable in the way they treat different cameras. However, it should be said that with ahigh MP file, interpolation artifacts will be at a very fine scale regardless.
At high ISO, the DxO data does seem to indicate that the A900 is a poorer performer than the 1Ds3 and D3/D700, even when compared fairly by compensating for the scale dependence of noise. It seems that Sony has some work to do in reducing the electronic noise of the sensor, in order to catch up with C/N. I've not done any tests myself, but the DxO data seems to indicate that the S/N performance is poorer than the competition at high ISO, and better than the competition at low ISO (one might speculate that this is due to the column-wise parallel processing of data coming off the photosite array). Since dynamic range is determined by S/N performance, that too is worse at high ISO than the competition and perhaps a bit better at low ISO than the competition, as Michael seems to be observing.
Bits -- the camera doesn't need 14 bits! According to DxO, there are only 11.5 stops of DR maximum, so 12 bits is ample. This will be more and more true as pixel counts grow, since the pixel level DR will go down as the pixel level (but not image level) noise increases; see above. Less DR means fewer bits needed to encode it.