A man simulating a woman is not a woman
If a man were to win the imitation game, he would only have succeeded in simulating a woman. He would, obviously, not be a woman. Therefore the imitation game is not an adequate test.
The argument as anticipated by Copeland (1993):
"Suppose the man passes the test (that is, forces a wrong identification frequently enough over a number of trials). Obviously this doesn't show that the man is a woman, just that in conversation he is capable of giving a good simulation of a woman".
Copeland, Jack (1993), Artificial Intelligence: A Philosophical Introduction, Oxford: Blackwell, pp. 46-47.