Oh, that's right. It increases linearly with the size of the CPU instruction words in a dimension and the size of the microcode instruction words on the other dimension.
The entire exponential addressing space is not used, since there is an overall structure to both instruction sets.