Haha! Sorry, I forgot to mention this in my first post, and it's pretty fricking relevant. If you are exporting to 24-bit, you really don't need to dither. If you plan on going to redbook CD (standard 16/44.1k) you will want to dither when you cut down on your wordlength.
The reason: 16 bits of wordlength give you 65,536 possible values. Since sound waves consist of both compression AND rarification (postive AND negative air pressure), that gives you roughly 32,768 levels of amplitude in each direction (positive and negative), excluding a "zero" value.
24 bits gives you 16,777,216 possible values, 8,388,608 in each direction (excluding zero).
You can see how much more accurate 24-bit recording is. And, in fact, it is accurate enough that you don't have to dither because your ears aren't really good enough to hear the bad quantization (distorted crappy sounds at low levels) with 24 bits. That being said, if you are just exporting to 24 bits, then just turn your dithering off.
By the way, I only learned this stuff from taking a college class in electroacoustic composition (we had to learn quite a bit about our equipment). That was over a year ago, and so some of it might be a little rusty. If anyone can correct me on anything, please do so in the hopes that any poor advice from me won't destroy Eric's pristine audio recordings.
Cheers,
~Brent