The issue isn't if the meter fails to expose as you creatively desire, shutterdancer, but how to maximize SNR.
The expose-to-the-right people would say to not set -4 EV in-camera, but rather to manually expose up to the point where you risk clipping (thus maximizing the full dynamic range of your sensor) and then to darken in post.
Think of it this way - the best tested DSLR by dxolabs (talking about dynamic range) is the Nikon D3X. It has ~13 stops of DR. If you purposefully underexpose 4 stops in camera, you have purposefully discarded 30% of what the camera is capable of. Since the camera can only record 14 bits of information, and all your final images will be 16 bit (8 bit jpg = non-linear and is ~equivalent to 16 bits as far as dynamic range is concerned) by setting -4 EV in camera you have limited yourself to 10 bits of DR in the final image. If you expose to the right in camera, and only subtract the 4 stops in post processing you will be left with 12 stops of dynamic range (because you're working in 16 bit, not 14 bit, space at the time of applying the effect.)
If we're talking a more mainstream camera (D300-class sensor) the DR in camera is only ~12 stops and you're tossing a full third of what the camera is capable of by doing the exposure compensation in-camera vs in-post.
