I might be a bit dense today, and hence deleted my previous post after realizing something.

If I'm not wrong, the seed isn't extended by adding 'Electrum' to it. The salt is however, 'Electrum + passphrase' instead of 'mnemonic + passphrase'. If the seed can be used in the salt to produce a different 512bit output, wouldn't there still be a considerable increase in entropy as long as the ENT of the input < length of the output? I'm sure I'm missing something here.
Let me put it this way:
We simply have a key derivation function that takes 2 inputs, A and B. If A is created from a 128 (or 132) bits of entropy and B has 0 entropy (no extension word) then your KDF is deriving its keys using that much entropy. If B also has 128 (or 132) bits of entropy then your KDF is deriving its keys using A + B bits of entropy.
Additionally we can say that in order to brute force this to get the BIP32 seed you'll have to generate and check both A and B so the entropy size is A+B.