17.♕f3 was about the only move to prevent …g5-g4, followed by …h4-h3. But after 17…♖g8, White can hardly stop …g5- g4 anyway.
19.♗e3 ♘xg4 20.fxg3 hxg3+ 21.♔g1 ♘xe3 22.♖f2 ♘f5 23.exf5 gxf2+ 24.♕xf2 ♗xf2+ 25.♔xf2 …0-1 (49)
If all the pieces were to be exchanged, the pawn ending would be winning for White. However, 20.♖xc8 ♖xc8 would clearly be a premature start, granting Black the only open file. Thus, Capablanca used the Karpov stratagem (again, an anachronism, of course!) to gain time for doubling his rooks on the c-file.
Indeed, the position is actually pretty good, and Black even resigned here! Black has no decent way to prevent 18.♗xf7+ ♔xf7 19.♕b3+. For instance, 17… ♔h8 18.♕b1 or 17…♖f8 18.♕f5, and Black’s queen remains without good squares: 18…♕e8 19.♘g5 g6 20.♕h3 h5 21.♗d6, winning big material. So, 1- 0 it was, right away!
Black no longer has the possibility of …h7-h6, to drive this knight away. Now, White wants to follow up with ♕d1-g4-h4.
The somewhat awkward 13…♘e5 was better, preventing White’s plan, though White is still better after the persistent 14.f4 ♘5c6 15.♕g4.
Planning to follow up with ♗c1-h6, but other moves were equally good here.