zydeco: High-quality game with some unusual, difficult decisions.
I'm surprised Najdorf didn't triple Bronstein's pawns with 21.Bd4 Qe7 (or Qe8) 22.Bxc5 dxc5 23.Rd3 and then double rooks on the d-file.
25.e5 seems like the right move. I imagine Bronstein was preparing an exchange sacrifice: either 25....Bxe5 26.Bxe5 Qxe5 27.Bxa8 Rxa8 followed by ....Bc6 or 25....dxe5 26.Bxa8 Bxh3 followed by ....exf4.
In terms of anatomizing Najdorf's defeat, it's probably that he played unambitiously in the middlegame; accepted a passive endgame position with 34.Qxd4?!; and then got impatient with 43.e5?!
I would have had trouble resisting both 37....Rb2+ to drive the king back and 39....f5 40.Bd3 fxe4 41.Bxe4 Bxe4 42.Rxe4 d5 although white is probably in good shape after 43.Re5 dxc4 44.Rc5.
Pachman includes 27....Qe5! in Modern Chess Strategy to demonstrate the importance of good bishop v. bad bishop. Euwe includes the entire ending from 49....Rd2 in A Guide to Modern Chess Endings and seems to think it was played accurately (he doesn't suggest any improvements for either side).