I don't like KenPom's use of the term "luck." I think random variation is the better term. More skilled players make a higher percentage of their shots than less skilled players, but there will always be variation because people's actions just by human nature will have variation. Sometimes a coach draws up a good play, and the player misses an open shot. Sometimes a coach literally doesn't call a play and just gives the ball to a player and tells everyone else to spread the floor, and the player makes a great 1-on-1 play and scores.
I'm guessing if you compiled a bunch of end of game data and ran a logistic regression model to predict the probability a team wins down 1, with ball, end of game situations using the offensive coach's career win % as one of the explanatory variables, that variable's regression coefficient wouldn't be significant.