Stan Wagon's March Madness Puzzle

I have been on Stan Wagon's wonderful problem of the week email list for some time but I confess that I haven't given any of the problems serious thought until now. I received this very timely problem a week ago and have been a little obsessed with it since. The problem "1358: Seedings and Probabilites", submitted by Erich Friedman of Stetson University in Deland, Florida, is given as follows:

Suppose there are 8 teams in a single elimination tournament, seeded in the usual way from #1 (best) to #8 (worst). In the first round, #1 plays #8, #4 plays #5 (and the winner of 1 vs 8 then plays the winner of 4 vs 5), and #2 plays #7, #3 plays #6 (and the winner of 3 vs 6 then plays the winner of 2 vs 7). The two unbeaten teams after the second round play in the final game to determine the tournament winner.

Suppose the lower seed (the better team) wins with fixed probability p, where 1/2 ≤ p ≤ 1. The #1 seed would prefer p=1 as that guarantees that they win every game. What value of p would team #2 want so as to maximize its chance of winning the tournament? Same question for the other six teams.

Now I'm far from done chewing this over, but I want to share what I've managed so far. I took a very computer-science approach and found some recursive algorithms for computing all the seeds, brackets, and probabilities involved. I coded it up in javascript, discovered the lovely JSX Graph, read up on the golden section search algorithm, and ended up with this little applet that solves the generalized form of the problem.

The x-axis is the fixed probability of the lower seed winning each game, and the y-axis is the probability that a given seed wins the tournament. Though using the sliders on the bottom, you can change the size of the tournament and the round to consider, to find the odds of things like a team winning its second-round game or making it to the semi-finals.

Graphing it all out I can see quite a few patterns, so I bet there's some more insight lurking somewhere. I want to analyze my recurrences to see if I can get something to pop out, or maybe I need to start over with a new approach. Once I feel satisfied (and probably once the solution is released so I can make sure I'm not completely embarassing myself), I might post a more mathematical write-up.