I did something similar with books[1], but drawing the words based on their F-score was a brilliant move. I think it gives the viewer much more information than word frequency, and thus is much more appropriate for generating word clouds. I wish I'd thought of it first.
Dunno, I think the raw frequencies work fine in your case, because there aren't really any themes that all (or many) authors across your selections keep returning to.
Yeah, probably. I'd have to do a hybrid weighted frequency/f-score thing, otherwise I'd lose some information, but yes, great idea with the F-score overall.