A different approach
I took advantage of Greg's work; the shots he made are excellent to this purpose. However, I have a different approach of comparing the dynamic ranges.
1. I am comparing the unadultered raw data directly (no demosaicing, no nothing). The question is, what the camera delivers, not how good the raw processor is.
2. I am using only two images from the serie, which are directly comparable.
3. The highlights play no role in this question, for
a. the raw data is linear,
b. what is clipped is not within the dynamic range, therefor "recovery" is nonsensical in this context,
b. what is captured is naturally within the dynamic range limited only by noise.
4. I am comparing in the darkest shadows the standard deviation of pixel values on some smooth, uniform areas and the appearance of fine details.
I selected the f/10, 1/250s shot form the D3 and the f/7.1, 1/500s shot from the D300, because
a. their exposure is virtually identical,
b. both have really dark areas containing smooth, uniform spots as well as fine details.
Here are the raw histograms, first the D300, then the D3; they show, that the difference between their effective exposure is less than 1/6 EV:
I located two spots, which appear uniform in color and smooth (from this distance). I made a selection in those areas and compared the pixel values as well as the standard deviations (which is the indicator of noise). This is easy, because the numerical pixel value range is the same in both cameras, namely 0 to roughly 16300.
The captures are greenish, because there are two green pixels for one red and one blue. Again, this is not demosaiced.
A small orange recangle shows the selection. The average pixel values within that selection are marked with a yellow disk, the standard deviations with a magenta disk. The two other numbers in the groups preceding the average are the minimum and maximum pixel value in the selection. There are separate values for red, green and blue (the two green channels are lumped together).
The first selection, D300 followed by D3. These captures show, that the standard deviation is much lower with the D3; it is obvious from the capture, that the spot is much smoother (well, this has to be so with the lower standard deviation). The values show, that this spot is in the ninth stop downwards, i.e. the D3 has a quite clean 9 stop, while the D300 is too noisy there (the ninth stop is from 64 downwards).
Now, let's goo deeper. The next selection is under the previous one, on the even darker area, which is at the end of the night and the beginning of the tenth stop. The D3 is
much cleaner than the D300:
Finally, a more subjective but not less important aspect: the reproduction of fine details. Keep in eyes, that this area requires +4 EV in ACR (with black = 0) to be this bright. This is 4.5 EV (I am not sure if it is known here, that ACR applies +0.5 EV autpmatically to these images, without indicating it).
Again, the D300 followed by the D3:
Anyway, I think it is indisputable, that the D3 is much better.
(I do not own a D3, nor a D300.)
Greg, thanks for the shots.