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".
Perhaps I didn't word that very well. I'm not saying that the 14-bit converter will give 2 stops better ISO performance, and I agree that the extra bits are at the "bottom" of the range. Let me try again with an example....
Set the 14-bit camera at ISO 100 and the 12-bit camera at ISO 400. If the exposure is correct for ISO 400, both cameras will give the same signal-to-noise raw outputs with the same number of useful quantization levels with the top two bit of the 14-bit camera always 00. However, the shot of the 14-bit camera will not be over exposed if the "correct" exposure happened to be ISO 100. So in effect you can shoot ISO 100-400 exposure at ISO 100 setting and get at least a good 12-bit output. If the exposure level is ISO 400, the 14-bit camera will not give a better output than the other one. If you happen to over expose what you thought was an ISO 400 exposure, you're protected by a couple of stops. I think the high dynamic range mode in the 1DIII is doing something like this.
What's more interesting is that you can think of exposing the bright parts of you image at ISO 100 ( the sunball for example ) and dark parts at ISO 400, and bring the ISO 400 parts up to ISO 100 in post processing. The signal-to-noise of the dark parts will be no better than an ISO 400 exposure, but it would not be limited by quantization problems.
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.
As I mentioned before even if the sensor has less than 12 stops of range as defined by the RMS noise level, there is still information below this level. It's better to record it than to cut it off, especially in cases where you have to bring up the darker levels for dynamic range reasons.