What is remarkable about the high ISO results from the D700 and the D3 is that they maintain very high saturation up to ISO 3200. This is in contrast to e.g. the 5D (according to comparative images I've seen) where the saturation goes down quite a lot which reduces color noise (of course). Basically Nikon is able to retain very low noise _and_ high saturation. I am not sure how this is found in the dxomark scores.
I guess this is not in dxomark scores, for there is no such thing. The color reproduction of the sensor is totally independent
of the ISO.
Nikon changed the pigments with the D300 and D3, increasing the efficiency of the "red" pixels. This created lots of problems: highly saturated colors have been captured, which could not be reproduced in sRGB. I analyzed quite a few cases, when the photog was complaining about "early red pixel saturation"; in fact, no red pixel saturation occured in any of those cases. The problem was the color space.
Nikon must have noticed the mistake, because the D90's sensor has different spectral response: the reds went down.
Following captures show the average pixel values of the raw channels on the "pure red" patch of a Gretag color checker; 5D2 ISO 100 and 1600, D3 ISO 100 and 1600, D90 ISO 100 and D3X ISO 100. It is obvious, that there is a huge difference between the 5D2 and D3, and that this change has been reversed with the D90. The D3X followed the D90. Furthermore, it is obvious that ISO plays no role here.
(The proportion between the red, green and blue channels should be observed. The values are "normalized" at 100 on the red, for easy comparison.)