Being a complete noob with regards to multiplayer (but with a couple of thousand skirmish games), I thought I would contribute as I best could to the balancing discussion.
Several places I read that there was need of a simulator.
"How hard can that be? There is already AI in the game..."
I thought that an evolution style tech tree balancing, where the tech tree mutates into a number of small variants, a large number of games are played by AI with different tech preferences, and only the most balanced variants of tech survive to next generation of mutations.
This would over some hundred thousand games give more balanced tech trees. And we could crowd compute the thing, like SETI some years ago.
But then I realized how big the domain is. AI choices (e.g. Early harrasment) x AI tech preference ( 10 % take cannons, 10% take mg, etc etc) ^ 2 (because there are two oponents) x minute tech tree mutations (increase weapon n ability y with 1%) x terrain situations (open terrain, closed terrain, strong defensive terrain, etc) x resource availability.
So my gut feeling was that the domain became too big to simulate, with many millions of full games necesary to get good results.
That lead me to believe that it is impossible to make truly balanced variations based on skill and flair.. The domain is simply too complex as describe above The system is prone to have imbalances introduced due to ripple effects.
So a simplification of the domain would seem to be a prerequisite to simulate, and maybe also to design balanced.