Out of curiosity, I started looking back at recent playoff series where one team had a better run diff and the other had a better W-L. Apparently pytha/run diff is no better at predicting series outcomes than overall W-L record.
If I had to guess, it might have something to do with the importance of a good bullpen in the postseason, which is also what tends to help teams overperform relative to their run differential.