Exactly. Testing for balance is a problem that can be done in polynomial time (O(n^2) for straight up unit type vs unit type matchups, a higher exponent if you want to test how utility abilities (repair, attack buff, etc) from other units can influence the battle, but that is still polynomial)
What I am saying is, AFAIK, finding a balanced solution for unit stats and abilities is fundamentally really hard, NP-Hard hard. As in "it would take years for the computer to give a good answer for a small scale RTS if you are very lucky" hard.
With some nifty hueristics and a good seed (such as an almost balanced game), I can see that a computer can help to tweak things to make balance better some, and even do it in a reasonable time frame. The hard part, the NP-Hard part, is getting that good seed in the first place.