Thanks very much for the bug report! I can fix this by laboriously protecting sc->t3_3 (an internal argument list) which means I'm stepping on it somewhere -- haven't tracked it down yet.