I remember a response to this which goes something like - when you have eliminated the impossible, what remains may be more improbable than having made a mistake in one of your earlier impossibility proofs.