And to be more explicit, quoting the paper in case someone really has doubts: «we deem a captcha scheme broken when the attacker is able to reach a precision of at least 1%». With our FancyCaptcha we are/were at 25 % precision for attackers, so yes, it's officially broken, and it's been so since more than a year ago http://thread.gmane.org/gmane.science.linguistics.wikipedia.technical/56387 so surely spammers got better in the meanwhile.
Nemo