Why isn't Monte Carlo algorithm used in crazyhouse?

[velocidrom]

If the game tree of crazyhouse (chess variant) if greater than the one in chess, why isn't Monte Carlo algorithm used for crazyhouse engines?
It has a great success in go (the board game) and I don't see why it would fail for crazyhouse.
One would simulate lots of random games from a given position, and then choose the move that leads to a higher win percentage. The advantage in crazyhouse over go for it is that in crazyhouse like in chess, it's extremely easy to check out who won the game. While in go even the top programs still do not count as well as the average player, for some position bots fail (even the 6 dan level bot "zen" on kgs go server) and thus may lead to innacurate moves.

[H.G.Muller]

I don't think MC is very helpful to combat the problem of tree size. It is mainly a remedy for not having good evaluation heuristics for non-terminal game positions. So you play to the end, and only evaluate win/loss there. As in most games this is obviously impossible even with very modes branching factors, sampling rather than exhaustive search becomes a necessity, meaning you take a big hit in accuracy. But if the quality of your eval heuristic was even worse, you still come out at a profit.

But for Chess-like games, even with drops, I don't think eval heuristics are that poor. So it wouldn't pay off.

[velocidrom]

Ok I understand! Thanks a lot for the reply.