Well, the problem with PBKDF2 is that it can be implemented with a small circuit.
This means it can be bruteforced at a fast rate using an ASCI (or even a GPU).
However, given that the keysize is way too big to even bruteforce a small portion of available keys, there are no practical security implications whatsoever.
If the number used to create the mnemonic is random (and not generated by a shitty/faulty PRNG) there is no negative effect from the key derivation function.
Gregg Maxwell comment "Effectively BIP39 is a thinly veiled brainwallet scheme with a woefully weak KDF. It's prone to misuse, and when misused it picks up all the bad properties you might expect it to pick up.)
This is completely quoted out of context.
BIP39 has nothing to do with a brain wallet. Maxwell was referring to the use of a password to additionally protect the seed.
If the original mnemonic code is known, it basically just is 'guessing' the correct password (which basically means that this layer of security is similar to a brain wallet).
If an attacker has the mnemonic code, he can simply bruteforce the passwords very efficiently (because of PBKDF2).
He was explicitly talking about the deniability in this context.
Practically, BIP39 is secure. It all depends on the RNG used.
If your seed is generated randomly (which then is being encoded into the mnemonic), you are fine.
Further, the 'plausible deniability' is not as strong as people think it is.
You can safely use BIP39 for cold storage, or you simply create a wallet using core (completely air-gapped of course), generate a few 100s or 1000s of addresses and use them to receive funds.
There are multiple approaches for cold storage.