Brute forcing ciphertext an offline attack, while attacking the authentication tag (the result of the CMAC-calculation) is an online attack: it requires an active oracle. That is: you can try and break the confidentiality of a message on your own PC network of computers. However, you need to send many ciphertext (including authentication tag) to a receiver to see if a guessed tag is correct or not and an invalid ciphertext is accepted (to test if a message is successfully forged). Generally you only get one oracle, and the speed of this oracle is usually limited (e.g. by networking speeds).
If an authentication tag is not accepted then that doesn't give any information to an attacker if the next message changes. This would for instance be the case if a unique sequence number is included in the messages (although usually the authentication tag is verified earlier, so oracle attacks would be possible). Furthermore, many protocols will break a session and re-establish the session keys if an incorrect authentication tag is received.
In conclusion: attacking an authentication tag is much harder than breaking a cipher, even if the order of the attack is the same. Because of this it makes sense to increase the key size even if the tag size stays the same.
If you need more than 128 bits of security is of course another question; 128 bits seems plenty as long as quantum computers are not available. 256 bit encryption should be considered for messages that require long-term protection.