As long as the thing samples fast enough and its A/D converter has enough steps, then accuracy only comes down to calibrating the thing.
Until reading this post I didn't know what a Wii was, but I've formed an idea from all the discussion. As a games device it would have to have reasonable resolution or it would be annoying to use in a game. (I also never play computer games btw). +/- 20% accuracy has been mentioned. That may be so, but I'm pretty sure it would be consistent and linear - that is the nature of electronic devices. For a cheap item that is just for games, it doesn't matter if it is 20% different than another one in the shop, as long as it is consistent and smooth.
Calibration couldn't be easier - while hooked up and recording as per klee's instructions, just hold it level on its base (ie, rest it on the kitchen table) for a few seconds, then repeat for face down, left edge down, right edge down, nose down and bum down. Each time you are applying 1G or -1G. You can then compare this to your saved data in Excel and work out the appropriate corrections.
For a correction factor (one you multiply) you compare LEFT minus RIGHT to 2G. For a zero correction (one you add or subtract) you compare LEFT plus RIGHT to 0G.
I know I am telling some people how to suck eggs and others won't know what's going on, but hopefully this is useful for someone. Maybe klee can incorporate the calibrations into his/her spreadheet.
PS I found this post via http://www.autospeed.com.au. I used to read the 16V forum religously - that's why I could log in to respond.