You are right Greg. I compared MasterDark to the noise-reduced MasterDark and although the mean for the image has only increased by 1.2 ADU, values for hot pixels were lower by 26 ADU (out of 64000).
However, the standard deviation for the entire Dark went down from 5.5 to 3.2, making the modification, IMO, worthwhile.
Hot Pixels go away anyway with Pixel Rejection algorithm in PI. I only started using Master Darks for calibration of Flats and, more importantly, for getting rid of gradients introduced by the camera in longer subs. I have not actually managed to notice these gradients in the lights, but since I know they are there buried in the signal, so I like to try to remove them
EDIT: On a second thought I think that slightly lower values for hot pixels won't matter, since we have a range of ADU values for hot pixels in a Master Dark anyway, some saturated, some not. When dark subtracting, I believe than normally hot pixels are replaced in the calibrated subs with average value of the surrounding pixels - otherwise we would end up with black holes in the calibrated images when subtracting saturated hot pixels.