Recently it was conjectured that an ElGamal-based public-key encryption scheme with stateful decryption resists lunch-time chosen ciphertext and leakage attacks in the only computation leaks information model. We give a non-trivial upper bound on the amount of leakage tolerated by this conjecture. More precisely, we prove that the conjecture does not hold if more than a View the MathML source fraction of the bits are leaked at every decryption step, by showing a lunch-time attack that recovers the full secret key. The attack uses a new variant of the Hidden Number Problem, that we call Hidden Shares - Hidden Number Problem, which is of independent interest.