Irock: Watching people play is important, but maybe that's more for testing how they learn to play the game? Or how they deal with the GUI? Rarther than to playtest balance?
I believe that to be true. I would also suggest creating a stats/balance tool that you can use to analyze balance data that you spit out to a file from your game. This allows you to have a bunch of playtesters whom you don't directly interface with, but create a large pool of balance information for you. In a fighting game, you could track which characters they use, how many of each move they use for those characters, and whether or not they win(along with how strongly they win...and what character they were playing against). This may sound complicated and perhaps like overkill, but this was an invaluable resource for us balancing Break Blocks, which has gotten a lot of positive remarks for the balance/progression in reviews.