After some days of work, I managed to add a simple captcha to Typo blog system. In my opinion, capchas like mine are more efficient at the moment of reject a bot. This is because the captcha is based on logical questions that the user posting a comment needs to answer. So, the answer to the captcha cannot be obtained from the text itself. Now, a better implementation of this would be using images with a little distorded text, but maintaning the question/answer aproach.
One problem with this would be the quality of the questions. Maybe some user would fell confused with trivial questions like "What color is green?", expecting that the answer will be something different from "green". Another problem relays on the answers. Some questions have more than one answer and it gets a little complicated, not imposible, to manage all expected answers.
My captcha aproach is now under test. I had lots of spam comments before implement it, and now that it's working, it seems to work really well. There is no way (I hope) to get the answers from the user data. A cracker needs access to the root files to get them, and that means cracking the entire server, something a spam bot will not do. There is no information stored on visible variables, only the question. So I hope it last as long as it can.