That's an excellent point to make, Emil. However, the DXO results would suggest that neither the P45+ nor the P65+ is using sufficiently clean electronics to match the lower loss of DR, as ISO is increased, that is apparent with the D3X. How do you get the impression that this is not the case?
For example, at base ISO, the D3x has a full stop better DR than both the P45+ and P65+ (slightly more than a full stop, but let's not quibble).
At ISO 800, the D3X has a full 2.5 stops greater DR than the P45+. Nikon's analog, pre-A/D boost, is having the desired effect.
Here's the DXOmark link to the graphs I'm looking at.
Edit:I forgot to mention for the benefit of those who never get past the single figure "Overview". Click on the 'Dynamic Range' heading.
I didn't say their DR was better at base ISO, I simply said that they are getting close to the full DR of the sensor they are using.
The ISO chart indicates that there is no analog gain for the P45+ beyond ISO 100; that is what the flat ISO graph means. The P65+ ISO rises from 100 to 3200, indicating that it does have hardware gains for each of these ISO's. So let's concentrate on the P45+.
DxO had a choice to make in reporting the DR of the P45+: Do they report the actual DR, max signal level over read noise? This does not change when the "ISO" is set above 100, since the ISO is metadata and the same analog gain as ISO 100 is used. However, a photographer would typically double the ISO to 200 when the ambient illumination level drops by two, so middle gray is going to be located at half the RAW level for a typical ISO 200 exposure compared to a typical ISO 100 exposure. The engineering DR at ISO 200 is the same, but only by virtue of there being an extra stop of highlight headroom above middle gray of typical exposures; the amount of "footroom" in shadows below middle gray drops by a stop. In other words, what changes is the typical placement of middle gray in the available DR. So DxO could report the engineering DR, but that would be misleading for photographic purposes, so I think they decided in circumstances where the ISO is done in software to simply lower the graphed DR value by one stop for every stop increase in ISO above the highest hardware-based ISO.
Now, the CCD in the P45+ is a 12-bit device. The DR of the P45+ at base ISO is quite close to 12 stops. This says that they are getting just about everything out of the sensor that could be got, and therefore there is no reason to have hardware gain. I'm not sure what bit depth of ADC they are using, though the output files are 16-bit, they may be oversampling the DR to eliminate quantization error, which would become more apparent as the RAW values are multiplied digitally to generate "higher ISO" if they didn't have enough bit depth.
Now, the D3x is another story. The sensor has more DR than the camera is delivering, and so it makes sense to have hardware based ISO gain. Here's how it works for my Canon 1D3:
Here I have plotted the S/N as a function of absolute exposure, for various ISO. At the upper end of exposure, each stop increase in ISO pushes another stop of sensor data past the range of the ADC, and so a stop of highlights is lost. But because the amplifier isn't clean enough at low ISO, increasing the ISO gain boosts the signal relative to the amplifier/ADC noise, and so shadow S/N improves with increasing ISO up to about ISO 800, and a bit more at ISO 1600 (I didn't plot the ISO 3200 curve since it lies on top of the ISO 1600 curve, apart from an extra lost stop of highlights). Thus the DR curve of the 1D3 is relatively flat from ISO 100-400, because although each bump in ISO drops a stop of highlights, the decrease in amplifier/ADC noise relative to signal adds back in almost a full stop at the shadow end. If the amplifier/ADC noise were less than the noise of the photosite array, then the latter noise would dominate, it would be amplified as much as the signal with increasing ISO; shadow S/N would not improve with increasing ISO, and DR would drop in proportion to ISO when using analog gain. And then there would be no point to having analog ISO gain; simply have an ADC with sufficient bit depth, and do ISO in software. You're not going to get less noise than the photosites themselves are generating, so why throw away potentially useful highlight data by having variable ISO gain in hardware?
Note that if the amplifier/ADC were less noisy than the photosite array, the S/N curve would be the envelope of all the above S/N plots -- that is, it would look like the ISO 1600 curve at low exposure, and the ISO 100 curve at high exposure. The full DR would be about 14 stops. I did a little analysis and simulation of what the images from such a camera could look like:http://theory.uchicago.edu/~ejm/pix/20d/te.../noise-p3a.html