I don't even understand why you talk of CS collision... CS collision is sure: given a known CS, there are about 2^248 combo (2^256 / 2^8) that give the same CS.
The reason I brought it up is that obtaining the second half of the mnemonic with a hypothetically larger seed would reduce its usefulness (while making the first half more useful), but then I realized that the CS is not an independent variable which can be modified without changing the entropy.