Ordinary implementers aren't doing de novo implementations of AES, and the gap between the P-curves and Curve25519 has closed, so this feels like a critique that might have been more germane 10-15 years ago?