not that i know much about this, but i'd say that it's most important that the SNR of the areas of each stack that will make it to the final image is adequate.
if you think of something like M42, you may only need 1-2s for the core in each subexposure, and even after stacking 50 images you're talking about a total integration time in the 1-2 minute range. for the outlying areas you may need hours of total integration time to get the SNR of the dust to a reasonable place.