Watson would likely lose if this was instead of jeopardy, a simple trivia test where contestants were scored for the most correct answers. Thinking about that really calls the contest into question for me. We let the machine off the hook for things people are good at but it is not (understanding speech, reading and vision) but we make no allowance for a machine's ability to press a button faster than humanly possible. If you instead phrased the contest as "machine presses button faster than humans" it doesn't seem impressive.
Your implicit assertion that natural language processing is something "a machine is good at" really calls the rest of your post into question for me. In particular, your demand that the researchers solve half-a-dozen Hard Problems instead of only one seems mildly mean-spirited.
