Maybe I've misunderstood, but it reads to me like you're saying that a 12-word seed offers equivalent security (~128 bits) to that of the individual addresses generated from it, and that therefore, longer seeds are overkill.
In Elliptic Curves the key's security is half the key size and since bitcoin key sizes are 256 bit that makes the security 128 bits.
The security of an entropy is equal its length, which means a 128-bit entropy is providing 128 bit of security.
Isn't that reasoning a little shaky? I mean, the seed is used to generate a unique sequence of addresses [1], no? Putting aside the increased hassle of longer seeds, isn't it desirable for it to be harder to recover the sequence than it is to brute-force a single address?
No because security is defined by the weakest link not the strongest. Meaning if the key itself has 128-bit security, it won't matter if it were created randomly or deterministically using a much bigger entropy. The key still provides the same security level.