One idea to counter the #wordscrambler bots is to see if they post faster in the replies than any human could type.

Another would be to check common words with main message, that ratio shouldn't be painfully near 100% for longer notes.

Also I just realized that optimizing spam protection, does out evolutionary pressure on bots to pass the turing test to such a degree we don't notice the spam anymore; and then... what's the point xD