Is that as far as noise is concerned or just in general? Then what does 14 bit get you?
First, the difference between 12 and 14 bits is that 14 bits allows you to specify more shades of color. With 8 bits (JPG), you can distinguish between 256 different levels of brightness for each of your three primary colors (red, green, and blue). When you go to 12 bits (older RAW), you can now distinguish between 4,096 different levels of brightness. With 14 bits (newer RAW), you can distinguish between 16,384 different levels of brightness. The range is the same (from completely black to completely white), but you have smaller steps between brightness levels.
The primary argument against 14-bit (and even 12-bit) is that you don't look at bits, you look at pictures on real devices. Computer monitors and printers can generally only show about 256 different levels of brightness for each primary color. The fact that there are more levels in your file doesn't matter because those levels get rounded off when you display the file.
That's pretty much true for cases when you take a picture and then do a straight conversion to JPG or directly print the file. The exception is for cases when you stretch the tonal range of picture. When you take a small slice of the dynamic range and stretch it, you can run into problems because you didn't have enough steps in the original. The result can be posterization or obvious jumps in brightness in what should be smooth transitions.
OK, now I'm going to paraphrase based on my quick reading of the article last night, so I could easily be wrong in my interpretation of it. The article contends that, for the cameras currently on the market with 14-bit RAW files, the extra granularity in brightness achieved by having the extra 2 bits is useless because the random fluctuation in brightness caused by noise is higher. In other words, you are trying to measure more precisely than the accuracy of the sensor. It's similar to the concept of giving poll results and a margin of error. Getting more precise than the margin of error doesn't really add any value.
So, if I understand the author correctly, while it may seem like you are getting more tonal precision because you are recording it with more digits of precision, you really aren't because the sensors aren't really that precise. So even if you buy they argument that there is a practical benefit to having 16,384 different shades instead of just 4,096, these cameras don't deliver that level of accuracy anyway.