just one comment - the master bias does not have too much to do with read noise. the bias signal is related to the very first step in preparing a CCD to generate photoelectrons - a positive bias voltage is applied to each photodiode. this way, when a photoelectron is liberated by an incoming photon, the electron goes somewhere and can later be counted. without the bias voltage eventually the photoelectrons would just randomly recombine with the holes that were left behind.
the bias signal is essentially a fixed pattern that is a characteristic of the CCD. the true fixed pattern of the bias signal is what the master bias is intended to capture - since the read noise is essentially random noise, you are driving the SNR (bias signal : read noise) up by stacking bias frames in the same way you drive the SNR of your DSO photons (DSO signal : DSO shot noise) when you stack lights. i think there is also some random variation in the bias signal itself, so by stacking you are also homing in on the true value of the bias signal.
in fact, if you subtract a master bias from an individual bias frame, the result is an image of the read noise in that single bias frame. essentially the read noise is always present in every frame generated by the camera, and can not be removed. that's the reason why exposure lengths are chosen such that the exposure is 'sky limited' rather than read-noise limited. you're trying to get the actual signal you're capturing up above the read noise threshold of the camera so that the DSO signal is not destroyed as the chip is read out. by stacking the bias and dark subs you're just trying to avoid injecting more noise than necessary when calibrating the lights.
rob
edit: clarification of the nature of the bias signal
