Bishop Pair v Knight Pair

EVERY serious chess player knows the bishop-pair usually conveys an advantage, or at least the prospect of an advantage to come.
Sometimes people joke about having the knight-pair, but can the knight-pair sometimes be an asset?


Black to make his 27th move in Spanton (1923) - John Wager (2029), Guernsey Round Six

A computer-run database study by IM Larry Kaufman - now a GM - showed the bishop-pair to be on average worth half-a-pawn more than bishop-and-knight or two knights.
In the diagram position, the knights are favoured by there being 14 pawns on the board, but the bishops benefit from the position being an endgame with no major pieces.
My main analysis engines, Komodo9 and Stockfish9, at first think Black has a small edge, but the more time they have to look at the position, the more they like Black. K9 comes to put Black the equivalent of virtually a pawn ahead, while S9 reckons the advantage is nearer a pawn-and-a-half.
27...h5
Black wants to open lines for his bishops.
In the post-mortem, JW thought 27...f6, with the same idea of opening lines, might have been better, but the engines prefer the text.
28.gxh5
I did not consider 28.Nh2 for long - such a passive move would almost certainly be worse.
28...gxh5 29.h4 Bf5
The light-square bishop eyes White's a pawn. If the latter advances, it becomes vulnerable to the dark-square bishop.
30.Kf2 e6
White's c pawn is also vulnerable, to the manoeuvre ...Bf8-a3-b2.
31.Ng5
Aiming to block the light-square bishop's action by playing e4.
31...f6
Opening more lines.
32.exf6
Better than 32.e4? dxe4 33.exf6 Bxf6 34.Ndxe4 Bxe4 35.Nxe4 Bxh4+, when Black emerges a pawn up, as in the game, but with a good bishop rather than a bad one.
32...Bxf6 33.e4?!
This loses a pawn, but I felt might give practical chances as White gets knight v bad bishop.
Objectively better may have been 33.Kf3 or 33.Kg3, but the position still massively favours Black.
33...Bxg5 34.hxg5 Bxe4
I was more worried about 34...dxe4!? Komodo9 agrees, but Stockfish9 prefers the text.

The ending has reached a new phase. Black has traded the bishop-pair for the win of a pawn, but is left with a bad bishop (all the black pawns are on the same coloured squares as the bishop)

35.Kg3
White ends up in zugzwang after 35.Nxe4? dxe4 36.Ke3 Kg7 37.Kxe4 Kg6 38.Kf4 a3.
35...Kg7
After this natural move, both engines for a while think the position is drawn … but it isn't!
36.a3 Kg6 37.Kf4
No improvement is 37.Kh4, although Black needs to find the idea 37...Bh1 38.Nf1 e5!, eg 39.dxe5 d4! etc.
37...h4 38.Kg4 h3 39.Kxg3 Kxh5 40.Kg3 Kf5 41.Nf1 Bd3
During the game, I thought Black should have pushed his e pawn immediately, but the text is stronger.
42.Nd2 e5 43.dxe5 Kxe5 44.Kf2
44.Kf3 is simply met by 44...d4.
44...Kf4
The engines show 44...d4!? is still playable, eg 45.Nf3+ Ke4 46.Nxd4 b4! 47.axb4 a3 48.Nb3 a2 49.Ke1 Bc2 50.Na1 Kd3 51.b5 Kxc3, and the pawn promotes.
45.Nf3 Bc4 46.Nd2 Bd3 47.Nf3 Bc2?
There was still a win by losing a tempo with 47...Bf1! 48.Nd2 Bc4, when 49.Nf3 allows 49...Ke4 while 49.Ke1 allows 49...Ke3.
The text hopes for 48.Ke2? Bd1!, but after …
48.Nd4
… Black has nothing better than …
48...Bd3
… which would allow a three-fold repetition, so a draw was agreed.