Captchas usually work fine for the most part, only in some instances it's hard to guess the captcha correctly.
As an example: Google uses pictures of house numbers. They gathered those pictures with their streetview cars and use the user input to get better results for navigation. (same goes for re-Captcha, which uses parts of books to recognize vague words)
Therefore the user only has to input some digits.
The approach from the comic takes the harder captchas and makes them easier by giving the user a task to type something. This will be considerably longer than those house numbers, resulting in many failing cases.
The other problem is, that captchas are still solvable by humans.
Example: Some shady porn-websites offer the user free access. The only thing they would have to do is login - after they correctly guessed the captcha. So they type the captcha in ... fail. Another captcha loads ... they type in, fail. Another captcha ... (and so on).
Because those captchas are copies of captchas from serious websites, that the attacker simply shows unsupecting users which they will solve for him, giving the spam-bot access to the serious website.
Therefore captchas are there to keep the bulk of non-smart spam-bots out, but are far from unsolvable. The usability must be high enough, so that users can gain entrance without doing too much and having too high of a failing rate.
If spam-bots are smart enough to sucess at every captcha, they simply might copy a nice, well-written and good-looking answer / comment from the same domain as the task and paste it. Often enough people dont read an answer in its entirety, but upvote it since it looks sophisticated.
Then were back at the original problem of spamming: bots post 1000 random comments and get entry with a small number of those. This might already be enough to earn the spambot-author money or entrance to a significant number of systems.