Steve Meyer, in his book Bishop v Knight, concludes the final chapter of the book, "Capablanca's Theorem – ♕+♘ is better than ♕+♗ in the Ending", as follows:
Verdict: Capablanca's intuitive insight into the advantage of queen and knight vs queen and bishop in the ending is correct. While general positional methods of evaluation are still important, the attacking force of the queen and knight can be very fierce, particularly in blocked positions or those featuring fewer pawns.
But John Watson, in his book "Secrets of Modern Chess Strategy", chapter "Minor-Piece Issues" claims that the examples are one-sided. He continues:
My own view, just from looking at a lot of examples, is as follows:
a) an unusually large proportion of ♕+♘ vs ♕+♗ games are drawn;
b) most games which are won by either side, as in the examples mentioned above, are characterized by that side having one or more rather obvious other advantages
Finally, and this might make the theorem a kind of self-fulfilling prophecy, he notes:
One proviso I have already noted above is that the side with the queen and knight (in the pure ♕+♘ vs ♕+♗ case) tended to be a bit stronger, for whatever reason. That alone might account for the small statistical edge for the ♕+♘. Moreover, as explained in the Introduction to this book, a perception of the superiority of queen and knight may well lead players who are in already advantageous but complex positions to convert them into what they view as a safer ♕+♘ vs ♕+♗ advantage. This would skew the percentages in favour of the ♕+♘.