atom was not speaking about the output length but about the details of the algorithm, i.e. whenever you have to apply the same hashing function thousand or even millions of times, the speed will be dropping accordingly (almost perfectly proportional). We can't make any direct conclusion/relations when we look at just output length and speed (the algorithm details and as said the iteration count is much more significant).
BTW: a fixed prefix within the mask is almost never going to give you the full speed. you might consider looking at other attack modes too (like -a 6 dictionary + mask). While this doesn't really matter too much for Veracrypt (slow algo), you shouldn't stick to the habbit to use a hardcoded prefix within the mask (especially for fast hashes like MD5).
BTW: a fixed prefix within the mask is almost never going to give you the full speed. you might consider looking at other attack modes too (like -a 6 dictionary + mask). While this doesn't really matter too much for Veracrypt (slow algo), you shouldn't stick to the habbit to use a hardcoded prefix within the mask (especially for fast hashes like MD5).