I know what you mean, but if you compare that to a list of the frequency distribution of *all* addresses, I'm betting that the distributions will be very similar - which means you don't gain anything by leaving them out.
And in fact, you're probably slightly slowing the generation process down.