The grey-world algorithm simply assumes that averaging all colors in an image results in a somewhat neutral value. Neutral simply means that Red =Green=Blue (or very close) and doesn't imply any specific value. The amount that your image varies from this premise determines the offset which is then used to mathematically "correct" the image data. In digital cameras, the green channel is usually used as a pivot point around which the red & blue channel data are modified. I believe Canon uses this method, but I'm NOT positive about that. There is also a white-patch method which is based upon maximum reflectance points.
Regardless...You are correct, good balance is when all three channels overlap as close as possible.
http://home.attbi.com/~ajpacheco/wb_test.htm
The graph's x-axis represents the six neutral tiles on the Gretag-Macbeth ColorChecker chart. Tile #19 is bright white and tile #24 is very black. There's about a five-stop difference between these extremes.
The y-axis is the RGB values read from Photoshop's Info palette.
For example, when I balanced on tile #19, the RGB values were very close to each other, at around 240 (0-256 scale). Using tile #22, which is virtually the same as an 18% grey card, the Red & Blue channels were starting to fall off. I tried this in a number of various lighting scenarios and had the same results.
Now is that a significant deviation?
Probably not for everyday shooting, but I'm currently looking into doing some camera profiling and white-balance is very critical in that regard.
How do I make sure the RGB values are greater than 200 in a shoot?
1)The grey card I use has a white reverse which has a 95% relectance (.05 density) just like the white tile (#19) on the Macbeth chart.
2) I usually have my camera's meter set to partial metering mode, a 9.5% area in the center of the viewfinder...I make sure the card covers that area.
3) I try to set the card so it won't be affected by reflected colors of any close objects.
4) I quickly focus on the card's edge, shift to the center, dial in an exposure compensation of + 2 stops, and make the exposure.
This will usually place the values up around 200+ with my camera/meter combination. You can also meter the card without compensation, make a note of the exposure, shift to manual mode, and add two or more stops to the exposure.
The idea is to get values that are around > 200 but not over 250 or so.
A quick look at your camera's histogram (info button) will show if you have any blinking areas (over exposed).
You can then use this image for in-camera custom white balance or perform the operation later using the conversion software of your choice.
I use BreezeBrowser but any Canon-based application will work fine.
BTW, I'm talking about processed images when I refer to these values, not the unprocessed linear data.
Canon's instructions seem to imply that you should expose the white target using the meter's results, which will produce an image more like a mid-grey. On this part I disagree.
Here's a sample of the chart's neutral tiles, first with auto-white-balance from MY camera...
And here with custom-white-balance applied...
You need to establish your own method. Whatever is easier for you.
