Perhaps as an experiment try "enter the first character of your post/comment". That would kill any remote human captcha solvers (since they don't actually know the post content), and likely require some rewriting of the spamming software, assuming that is automated.

Of course, this is not a long term perfect solution against motivated adversaries, but it's a way to see how the current spammers work.

Another: Add a keydown handler to your message-textarea and log (to a hidden form field) how many key presses are being used per post. If the spam software is setting the content field programmatically, you then know how to detect them.