Callix and JaronK, what build are we trying to get a probabilistic bound on number of attacks for? And Callix, have you tried computing the variance on it?
The basic build, IIRC, was Fighter 2/Warblade 10/Disciple of Dispater 8 with a pair of Aptitude Kukris, using TWF, ITWF, GTWF, Weapon Focus (Light Mace) Lightning Mace, Improved Unarmed Strike, Roundabout Kick, and Improved Critical (Kukri), plus the prerequisites for DoD. Because of the Aptitude enchantment, Lightning Mace and Roundabout Kick apply to your Kukris, so you critically hit on a 9-20, and every critical threat gives you an extra attack, plus every confirmed critical gives you an extra attack. As such, you've got a 55% chance with each attack to get an extra attack from the critical threat, and then if that triggers you've got a 95% chance to get a second additional attack, assuming you hit on a 2+ (which, with Blood in the Water, is extremely likely). Using Dancing Mongoose helps a bit as well.
Now, this particular mathmatical problem is already notorious for screwing with summations. A similar problem is the doubling quarter flip game. Let's say I make a gambling game, and I tell you that you have to give me some money, and then you flip coins. If you get tails, game over and you lose. If you flip and get heads and then get tails, I give you $1. If you flip and get heads, then heads, then tails I give you $2. If you get three heads and then tails I give you $4. If you get four heads and then tails I give you $8, and so on. How much money would you expect to win, and thus what's the most amount of money you should give me and still expect to come out ahead?
Well, if you do it the normal way, you have a 1/2 chance to win nothing. You have a 1/4 chance to win $1, a 1/8 chance to win $2, a 1/16 chance to win $4, etc... that comes out to 25 cents + 25 cents + 25 cents, etc all the way to infinity, suggesting that your average pay out is infinite, and thus you should happily give me $100 play my game, and if we play enough times you'll be rich indeed. However, the actual average payout is around $11, IIRC, if you just run it in simulations. It's a really interesting mathmatical thing.
JaronK