Worth A Punt (part two)?

AFTER 1.d4 e5!? 2.dxe5, the main move is 2...Nc6, which White usually meets by developing a knight of his own, while simultaneously holding on to the e5 pawn, viz 3.Nf3.
Black normally replies with the tricky 3...Qe7, but Emil Diemer of Blackmar-Diemer Gambit fame liked to sac his f pawn.
Edgar Walther - Diemer
Swiss Championship (Thun) 1956
3...f6!? 4.e4!?
The analysis engines Stockfish10 and Komodo10 strongly prefer this over the obvious 4.exf6. After the latter, one of Diemer's games continued 4...Nxf6 5.Nc3 d5 6.Bg5 d4 7.Ne4? (White has an edge after 7.Nb5 Bb4+ 8.c3 dxc3 9.Qxd8+ Nxd8 10.Nxc3) Nxe4!, when Black wins a piece. G Brachtl - Diemer, Baden Seniors 1978, saw 8.Bxd8 Bb4+ 9.Nd2 (9.c3 dxc3 does not help White) Bxd2+ 10.Qxd2 Nxd2 11.Bxc7 Ne4 (0-1, 37 moves).
4...fxe5 5.Bc4 Nf6 6.Nc3
6.Ng5!? d5!? 7.exd5 Na5 has a Two Knights' theme about it. White can hold on to his extra pawn with 8.Bb5+ c6 9.dxc6, but 9...Qxd1+ 10.Kxd1 Nxc6 gives Black some compensation, although the engines prefer White.
6...Bc5?!
The engines much prefer the pinning 6...Bb4.
7.0-0 Rf8?!
Almost certainly better is the natural 7...d6.
8.a3?!
White may have a small edge after this, but the move is slow. Better is the engines' 8.Ng5 or 8.Nd5.
8...d6 9.Qd3 Nh5?
Black more-or-less equalises with 9...Bg4.
10.b4 Bb6 11.Nd5 Bg4

How should White proceed?

*****
*****
*****
*****
*****
12.Nxb6
If 12.Bg5, Diemer may well have planned to sac the exchange with 12...Rxf3!? After 13.gxf3 Qxg5 14.fxg4 Qxg4+ 15.Kh1 0-0-0 the engines prefer White but the position is not clear.
Also unclear is 12.Ng5 Nf4 13.Nxf4 exf4 14.Nxh7 Ne5.
12...axb6 13.Ng5 h6 14.Ne6 Bxe6 15.Bxe6 Nxb4?!
Diemer consistently chooses the sharpest option, but the calmer 15...Nf4 16.Bxf4 exf4 may well have served Black better.
16.Qe2 g6!? 17.Bxh6 Rh8 18.Qf3 Qe7 19.axb4 Rb8!? 20.Bf5!? Rxh6 21.Bxg6+ Rxg6 22.Qxh5
The smoke has cleared, and White is a pawn up. But the mass of heavy pieces means it will not be easy to convert his advantage.
22...Qf7 23.Qh3 Ke7 24.Qh4+ Kd7 25.f4 exf4?
This helps White. Much better is mobilising his passive rook by 25...Rbg8.
26.Rxf4 Qg7 27.Qh3+ Re6 28.Raf1 Qd4+ 29.Kh1 Rbe8 30.Rf7 Kc8(??)
Did Diemer really play this? According to the score in ChessBase's 2020 Mega database he did, but much more sensible is breaking the pin on the e6 rook by 30...Kd8.
31.Qa3(?)
Much stronger, if …Kc8 really had been played, is 31.Rf8.
31...R6e7 32.Qa8+ Kd7 33.Qxb7 Qxb4 34.Qd5 Kd8 35.Rxe7 Rxe7 36.c4 Kd7 37.h3 Re5 38.Qd3
Stronger is the engines' 38.Qg8!, eg 38...Rxe4 39.Rf7+ Re7 40.Qg4+ Ke8 41.Rf1 Kd8 42.Rf8+ Re8 43.Rxe8+ Kxe8 44.Qc8+ Kf7 45.Qxc7+, although queen-and-pawn endings are rarely simple.
38...Rc5 39.Rf7+ Kc6 40.e5?!
Stockfish10 likes 40.Kh2, but the rook-and-pawn ending after 40...Qxc4 41.Qxc4 Rxc4 is not clear.
40...Qe1+ 41.Rf1 Qxe5 42.g4
This gets a double-question mark in Mega20, but the move is fine (albeit the position is equal).
42...Kb7 43.h4?!
Still trying to win, but it was safer to take the draw, for example 43.Rf5 Qe1+ 44.Rf1 Qe6 45.Rf4.
43...d5! 44.cxd5??
This is the losing move. White had to play something like 44.Rf5, although Black is better after 44...Qe1+ 45.Kg2 (best) dxc4.
44...Rc3 45.Qe2? 0-1
45.Qe2? loses instantly, but 45.Qd2 Rh3+ 46.Kg1 Rg3+ 47.Kf2 Rxg4 is also hopeless.
This is an imperfect game, to say the least, but it illustrates some of the attacking chances a tactically-minded Black can generate.