> What percentage of games can one expect to win if one plays perfect
Well, we don't know of course but one could make some sort of an
educated guess. It depends on the skill of the opponent of course. If he
also plays perfect backgammon, than you can't win more than 50% assuming
it's not decided yet who gets the first move.
Based on FIBS rating system, and what ratings the best bots have
achieved there compared to the average rating, I estimate that for
simple money play, a perfect playing bot would have a rating somewhere
between 2200 and 2500. Let's say it's 2350. Against an average FIBS
player with a rating of 1550, we get:
Rating difference D = 2350 - 1550 = 800
Probability of underdog to win a 1pt match: 1/(10^(800/2000)+1) = 28.5%
It's not clear how matchplay and FIBS rating system for it translate to
money, but I think the winning percentage for the underdog is somewhere
between these values, and closer to the two point match, since money
games involve (back-)gammons and full cube usage.
So, my estimation would be something like "a perfect player would win
around 77% of the points (probably also games, but not necessarily) in a
money game against an average FIBS player, translating into an average
turn-out of 0.77-0.23= 0.54 points per game."