There would have to be a great deal of data and there would have to be data separated by winner breaks and loser breaks. Further, the breaker of each game would have to be known.Well it is destined to turn into that. Here is something you could do to see if your model correlates to actual data. Fargo Rate and one of the Wisconsin leagues tracks player performance through games win and lost. As such there have thousands of actual match results from both formats. It would be interesting to see if the actual results match up to the predictions in your model and Bob's....
The OPs simulation had 20,000,000 games between just two players.