I'm interested in how the odds change as tiles are played. When, say, the J and both Vs and Ws have been played, the remaining tiles are less dangerous.
I'm curious though by your characterization of this as being "game theoretic"... It seems like this strategy is a consequence of correctly analyzing the game state tree, which doesn't on the face of it seem to involve game theory to me (in the way that term is usually used in economics.)
They are treating this as a stochastic game. They are accounting for their opponent's likely response when deciding their own move. That is enough for me, anyway.