My battery died around 75% through the wordlist, even though most the number crunching was on a remote server. The software I was using goes into an extremely slow decode process when using correct password on test images other anons posted. Therefore you can tell if a password wasn't rejected immediately.
Instead of general vocabulary trials we should compile list of more probable passwords. The vocabulary of Q posts (including image file names) is one start. The vocabulary of the original 4Chan post likewise. Newspaper quotes could be a lead too. It should try one word at a time, then 2 following words at a time, 3 words in a row at a time all from the text corpus…
I don't understand the F5 algorithm but if wrong passwords are rejected quickly most the time then a timing attack could guide the crack, maybe. But almost all trials quit early anyways.