[RiderOfGiraffes]

  > For N balls, the other balls can/should be binary searched.

Interesting assertion. Can you prove that? Can you give a provably minimal search pattern?

  > It's just a little tricky when it comes to the last two.

Another interesting assertion. It seems to me that the case of exactly two balls is trivial, once the solution is found.

[OmniLarry]

Sorry, can't find the link, but it was a Topcoder problem at least 5 years ago.

Well, it is trivial with two balls, once you get it, but it's tricky with respect to "just binary search".