If either of the above cases applies to your application, then you should absolutely be using a committing AEAD mode.
If you know your nonces will never repeat, would you sacrifice performance to use a mode resistant to nonce misuse? Of course not, this would be silly. If you are using a mode that can authenticate the ciphertext before decrypting it, would you throw away performance to use a mode resistant to the release of unverified plaintext? Heck no. If your keys are generated uniformly at random and are not under adversarial control, would you give up performance to use a committing mode? Absolutely not.
Always put forth due diligence when selecting a mode for your application. List out the exact requirements of your application, and select a mode that satisfies those requirements. It is perfectly safe to use a non-committing mode if you know you do not need it. Additionally, keep in mind that using cryptographic keys derived from low-entropy passwords will always be insecure, regardless of the committing property. The industry is actively moving away from passwords; once passwords are obsolete, it is difficult to justify the added complexity and performance overhead of committing AEAD.
The committing property is just one in a long list of existing properties (nonce misuse resistance, RUP resistance, etc.). My opinion is that the pendulum has swung too far in the direction of committing modes, and once passwords become less common, the pendulum will swing back to a more sane position. As always, put due diligence ahead of hype and FUD. Continue to avoid paying for features you do not need.