Hum..... this could be the source of the problem. “We plugged it into Fritz, the standard practice computer for chess players, which did three-quarters of a billion calculations, 20 moves ahead," explained James Tagg Co-Founder and Director of the Penrose Institute.... from
http://mashable.com/2017/03/14/solve-this-chess-puzzle/#bHEhjUAMmiqtwhich refers to the same puzzle.
A) Fritz, not Stockfish or Komodo?
B) 20 moves ahead? That's not enough for this position surely, hence why the eval remains at -x until a large depth is reached.
I'm even more confused now.... isn't this just another example of the horizon effect, or at least a demonstration that the current method of programming chess engines doesn't account for these few (very odd) positions?
"Chess computers fail at Penrose’s chess puzzle because they have a database of end-games to choose from. This board is not, Tagg and Penrose believe, in the computer’s playbook. “We’re forcing the chess machine to actually think about the position, as opposed to cheat and just regurgitate a pre-programmed answer, which computers are perfect at,” said Tagg."
Sounds like they are referring to endgame tablebases. Of course the computer can't do that, because a tablebase including this position doesn't exist. Why should it? 3 bishops is not really a common enough thing in games for this to exist. I've only seen extra bishops in compositions. Yes, there are legal games which can lead to this sort of position, but not practical ones.