Finding out a -rough- word count for a piece of literature should be fairuly easy. Indeed, searching for the regular expression /[^\s]+/ and spitting out the number of matches found should give you a reasonable accurate count of actual words (irrespective of length). A more complex algorithm would likely be preferable, but I don't know all that much about how word counts are usually calculated, to be honest. 

Certainly such a pattern match would be inefficient, but it only has to be done on edits, which is not hugely onerous.