My point was that seeds that offer higher security levels than the individual keys they generate are useful because they're protecting multiple keys.
But it doesn't matter how many keys you
generate from a seed, they all have the same security. Meaning if you generate 1 key using a 12 word seed it has the same security as the 100th key you generate from the same seed.
with a 256-bit seed being used to generate 64-bit private keys (with the top 192 bits set to 0, for example) the smaller 64-bit private keys don't reduce the security of the seed all the way down to their level.
The seed will have its 256-bit security (assuming 256 bit were generated and used) but the key will still have only 64 bits of security, not more.
In some sense, the seed "contains" the private keys that it generates, and viewed through that lens, the following example amounts to a very similar thing, even if it looks unrelated at first glance:
Imagine you have a file, listing the locations of secret military bases encoded as 64-bit coordinates (32-bit longitude, 32-bit latitude). Even though this file contains only a sequence of (sensitive) 64-bit values, that shouldn't decide what security level is chosen to protect its contents. It wouldn't make sense to argue that encrypting it with anything more than a 64-bit key is technically unnecessary.
That's a bad example because encryption is different and irrelevant in this context. When encrypting something, the message is not the deciding factor in choosing the algorithm, security level and key size.
But when deriving a key, the key size (or better say security level) itself is a deciding factor in choosing the entropy size.