Why roughly 128 bits, and not exactly?
Reading BIP39, 12 words seems to imply exactly 128 bits of entropy.
Why the additional entropy provided by using more than 12 words is not used?
This is nonsense; it is used, it just doesn't in exactly same way as the first 12 words. It's probably trying to say that more than 12 words does not meaningfully increase security, but in a very confused way.