The area of the 1.6x sensor is about 2/3 the area of the 1.3x sensor, so the signal-to-noise of the 40D should be within one stop of the 1DIII at the same exposure and ISO. Even if by the usual definition of dynamic range, max signal/noise level, the sensor is not quite 12 bits, there is still information below the RMS noise level and the extra resolution will still help. It'll make the parts below noise level fuzzy instead of banded.
I'm not sure what the raw, base ISO capability if the 1D3 sensor is, so assuming 1 stop less doesn't tell me what the 40D sensor will be. We know that the D200 sensor can achieve 12 bits and not any more, the 40D is a newer design but has to make do with a slightly smaller pixel. If Clark is to be believed, the D200 has at least of a stop of improvement available, so I could see the 40D sensor offering 13 stops. I think that would be remarkable though.
Looking at it another way, with a 14-bit A/D the parts of a shot that are 2 stops under exposed will have the same sensor output as the same sensor with a 12-bit A/D set to 2 stops higher ISO. Assume the same sensor, the output of an ISO 100 shot with the 14-bit A/D will be no worst than an ISO 400 shot with 12-bit A/D. If nothing else, it will give you room for underexposure in cases where you want to really be sure to preserve the highlights.
If what you are saying is that 14-bit gives you a 2 stop improvement in ISO performance, as would be suggested by your explanation, then absolutely not. If that were the case, all we'd need for great high ISO would be 20 bit converters!
The output of the sensor, after amplification, is referenced to a set voltage which is the maximum voltage the ADC responds to. This is independent of the number of bits in the converter. The "extra" bits added in the 14 bit converter, therefore, aren't at the "top", they're at the "bottom". The analog voltage representing 2 stops below maximum will be 1/4 of maximum voltage regardless of bit depth or ISO. If a sensor has less than 12 stops of range, then both the 12 and 14 bit converters will be digitizing noise in their lowest bits.
Assuming the same sensor with substantially less than 12 stops of dynamic range, both the 12 bit and 14 bit converters are capable of delivering the full range of the sensor and the 14 bit converter does not offer better performance at ANY ISO. There may be qualitative differences in noise in cases where the sensor performance is very near 12 stops as is the case with most DSLR sensors today. 14 bit converters may offer potential advantages at lower ISOs in that case. I would like to see a demonstration of that, however.
I don't doubt that the 40D could benefit from better converters or that it is capable of slightly better than 12 stops, but I would just be surprised if the benefit is substantial. The proof will be in the testing, and that proof will probably NOT come from dpreview.
It should also be noted that 14 bit converters are harder to make and can also be noisier than 12 bit ones. I have no doubt that Canon has a handle on that, but it makes the situation murkier still as a 14 bit converter *could* result in worse performance theoretically.
I'm not surprised that Canon put the 14-bit A/D in the 40D. It probably cost them less to make just one Digic III and associate IO that process 14-bit and use it for all the camera than to make 2 versions. The only incremental cost is a little more memory for the same shot buffer.
I think that's right and, furthermore, Canon gets mileage out of the branding of 14 bit. Just as Canon has convinced people that they are better because of the "Digic" branding (much like the "L" branding), Canon will take marketing advantage of their 14 bit converters whether they offer performance benefits at 1.6x crops or not.
