dcraw can't do this directly, but it can do something which will get you pretty close: render each pixel as a grayscale, either with unscaled pixel values with -D or scaled ones with -d. This will give you a PGM format image which will look like this (cropped and zoomed in, so each square is one pixel):

In this image, the bright dots happen to be the blue-filtered pixels. You can process this image in various ways to extract just those and discard the other. You'll have to, of course, decide what you want to do with the spacial discrepancy, because of course (this not being a Foveon sensor) the various colors are offset from each other. One easy approach is to just divide the whole thing in half in each direction (quartering the area).

If you want to do this in a less hacky way, try LibRaw, a C/C++ library which is in many ways the successor to dcraw. You could write a simple program to get the data you want and do whatever you want with it. That, however, is out of the scope of photography, really, and probably better on Stack Overflow.