Very good question here. Well, the Monte Carlo does not improve the strategies. It does not change them. What is does is to filter them.
Especially if you are using the Reactor and you include the Monte Carlo simulation, the strategies will be filtered. Only those that pass the Monte Carlo will continue.
Now, the Monte Carlo performs different tests for the strategies according to the settings we select to use. For example, when it changes the parameters of the strategy and it fails, this means that the original parameters are most probably over-optimized.
Also, if you test the spread and it fails, it means that the strategy is not suitable when trading with a higher spread.
For, example recently I create very nice strategies for Apple and I started trading with them. The EAs started to lose. It turned out that the broker charger $5 commission when the traders are opened and when they are closed. This means $10 commission for each 1 lot. That is huge!
Now, with the Forex, we do not have that problem. But the Monte Carlo simulation does pretty much the same thing. It tests the strategies with different spread, indicator parameters, backtest starting bar and so on.
In other words, it tries to break the strategy before we trade with it.