White is almost certainly going to play 38. a5 and not b3 forcing black bishop into precarious position trying to protect pawns in white square or risk a bishop exchange and passed pawn in A file. The knight will be tied up in its defense.
This is an interesting endgame. Clearly white has good winning chances but I am not yet certain of the correct result with best play.
38.a5 appears strong but after 38.a5 Nc7 39.Ne4 Bb5 40.Bxb5 Nxb5 41.Nf6+ Kf8 42.Nxg4 converting the extra pawn into a win will be difficult. Blacks defensive task has been simplified by bishop exchange. If all queenside pawns are traded then many resulting endgames with 3 pawns and knight vs 2 pawns and knight on the kingside will be drawn.
I prefer another plan which will keep material on the board while I centralize my king. Piece activity is very important. If possible I would rather win a pawn on the queenside than the pawn on g4. I am not certain that 38.b3 is a better move than 38.a5 but it may give me better practical winning chances.
Your move bitcointalk. You have the black pieces
The current position is updated below
Game 2 (letsplayagame vs bitcointalk):
1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Ba4 Nf6 5.O-O Be7 6.Re1 b5 7.Bb3 O-O 8.c3 d5 9.exd5 Nxd5 10.Nxe5 Nxe5 11.Rxe5 c6 12.d4 Bd6 13.Re1 Qh4 14.g3 Qh3 15.Re4 g5 16.Qf1 Qh5 17.f3 Bf5 18.a4 b4 19.c4 Nf6 20.Re1 Rfe8 21.Be3 c5 22.Nd2 Rxe3 23.Rxe3 cxd4 24.Ree1 d3 25.c5 Bxc5+ 26.Kh1 Rd8 27.Re5 Bh3 28.Qe1 Bd6 29.Rc1 Bxe5 30.Qxe5 Qg6 31.Rc6 Rc8 32.Rxc8+ Bxc8 33.Bc4 Bd7 34.Qd4 g4 35.Bxd3 Qg5 36.f4 Qd5+ 37.Qxd5 Nxd5 38.b3