At some point the 2nd team have better programmed players, that's what happen when the simulation choose your "worst combo" of players as the choosen ones to solve your match and, despite the AvQ diff, the simulation makes participe the best players of the lower AvQ. This more the sum that say that the simulations don't attending are more stable, did the rest.

Ive always said that the margin of beatability is arround the 30% avQ diff, but a 40 in that case seems not strange.
So the truth is that would be interesting to track this 11 nd see if still failing more often, I bet that yes, as in many cases when one buy just a 11 with quality without test the players to know how they fit, or if there's some player with negative impact -what I guess is the situation- .