Because the more complex decoder is traded in this case for a denser instruction set, which means they can trade it for less instruction cache (which is more power hungry).