At the moment, num_words is mostly a placeholder. PDFs estimate their number of words (something like 500 words per page), and the predicate will work for that. If I add html or any other type of text import support in future, I can add their number of words to the db for this predicate.
There is currently no way to discriminate a search by num_namespace. I like the idea, though. I'll think about it a bit–maybe I could modify num_tags to take a namespace?