I recently came across a discussion about automated reader of CAPTCHA images. A guy told they sold an implementation of such a reader for $100K (in total; certain initial payment followed by $5K/mo payments). While this might appear to be an exaggeration, I recalled another interview given by another OCR fellow who mentioed a simiar offer he declined for reasons he chose to not specify.
I am afraid I am losing something here, as CAPTCHA reader questions in in most cases not an issue as soon as it is required to decode particular type of images. Image prefiltering followed by OCR will pass through 95% of the protection implementations around, one need only an experienced software engineer and a desire to break the protection complemented by a modest budget. Moreover, CAPTCHA code can be changed anytime so the game is actually of a different nature: one makes it harder to decode in automated fashion and the other tries to get even. I would rather say that the task for the former guy is more difficult (as soon as we still expect web user to be able to recognize the code).