Here's the rundown, by move #, and according to Rybka (at fairly low depth calculation):

9.) 'Thinks' e5 was indeed better, as a few others mentioned here. Indeed, e5 is THE thematic break for white against the Benoni and those early variations where white forces through an early e5 (e.g. Taimanov) are undoubtedly one reason why the Benoni has lost popularity in more recent times.

13...) 'Thinks' Qa5 was better than Tal's Nc5, which took an eye off e5 and allowed Reshevsky's next move.

16.) Much preferred Nd2, presumably with the intention of targeting black's c pawn, and probably the d6 square ultimately. The text move just seems to weaken white's king side defenses prematurely.

18.) 'Believes' f5 was a bit better, although it ultimately prefers a5 over both. For reasons I'm somewhat unsure of.

21.) 'Thinks' this was one of Reshevsky's big mistakes in this game, much preferring Rf4, probably with the intention of supporting white's g-pawn, and his king side in general. Which Reshevsky's earlier g4 weakened considerably. Apparently Reshevsky didn't see his king side pawns as being vulnerable.

...21.) 'Thinks' Bxg4 was a lot stronger. Although the text move was Rybka's second choice.

...22.) 'Thinks' Rxg4 was considerably stronger than the text (which was also Rybka's second choice).

25.) 'Thinks' Ke1 was much better than what Reshevsky chose, which was Rybka's least favorite of white's three choices.

28.) 'Believes' this is essentially where white lost the game, as interposing any one of the minor pieces in front of the rook would have been considerably better than moving the king, which Rybka evaluates as about -3.00. Probably because the text move allows the rook skewer that Tal played on his next move.

29.) Much prefers Kc5, but basically thinks white is lost at this point anyway (-3.07 evaluation).

These were by no means the only imperfections in the play according to Rybka; only the most salient ones.