Rob, to expand a bit on what others have replied with.
ICNR = great...but it only deals with REPEATABLE NOISE. It doesn't deal with random noise in the darks (each dark and light will have a certain amount of random noise. Depending on temperature (see below), this may or maynot be an issue
Taking darks after and then applying them = could be great.... but often isn't. Stacking a multiple of Darks will deal with repeatable noise as well as reducing random noise by the sq rt of the number of darks you use to create a master. This is a very desirable thing. UNFORTUNATELY..... your darks, unless taken at the
same temperature will have differing amounts of dark current so combining them will not necessarily have the same dark current as your lights. In fact most probably not.
If there was say a 6 degree drop in temp between taking your lights (which will have different amounts of noise 'cos the temp is dropping anyway) and the final dark you taken, then there will be 1/2 the noise in the final dark as in the first light.
what does this all mean....IF you temp is stable (<2 or 3 degree maximum change ) then yes taking darks at the end can be a good thing, especially if you are pushed for light time. If though your temp is dropping (or rising) then consensus seems to be to take ICNR.
HTH
BTW non-random noise is not really noise by definition, it should really be described by something like unwanted signal or detrimental signal or something similar. jmo