It's abundantly clear that I'm not making a precise computation here. My argument is that tests don't scale linearly with the number of features because interactions between features need to be tested as well.