Speculation: the winning author is simply good at StarCraft and knew what to implement. His implementation was probably simple but did one thing very well, which beat more complicated/smart AI implementations from worse players.

Monte Carlo over the StarCraft state-space would be very hard to implement but I don't see any particular reason why it wouldn't work decently well.