They could revert to the old rule where you could buzz in as soon as you thought you knew the answer. It's hard to tell how long it takes Watson to come up with answers right now, but switching to the old rule would introduce an interesting dynamic where Watson would have to decide whether to spend more time chugging data or buzz in.

Also, this would be pretty artificial (no pun intended), but they could analyze previous all-human Jeopardy! episodes and figure out average buzzer response time, and perhaps incorporate that into Watson.