Mark, Andrew;
In playing with it, I noticed that the Scale slider that you mentioned seemed to be just a magnifying glass showing fewer or more pixels, and seemed to have no effect on the actual sampling size. Yet you quote Martin Evening as saying the sample size is dependent on the Scale. So to verify this, I decided to test the sensitivity of the cursor position to see if I could figure out how big the sampling size was, and if it was different depending on the Scale.
In Photoshop, I created a 30 pixel per side square in black. Then added a white 5x5 box in the center. Imported it into LR. In the Develop Module, I Zoomed up to 11x1 magnification, selected the WB tool, turned on "Show Loupe" and moved the Scale slider all the way to the left (greatest loupe magnification) and started moving the sampling tip across the box, watching the lab values and RGB values as I progressed, pixel by pixel. In the center pixel in the white box, L and RGB=100. That must mean that the sample can't be bigger than 5x5 (but could be smaller). Then I slid the cursor point into the adjacent pixel, and the L value dropped. That means it IS 5x5. It kept incrementally dropping when I moved into the next pixel till I had moved it five pixels right or left, up or down. L and RGB had dropped to zero in the 5th pixel. This confirmed that it is 5x5 with the slider all the way to the left.
Then I set the Loupe slider all the way to the right (least magnification) and repeated the test, pixel by pixel. Exact same results. 5x5 sampling area. So, my results showed that the size of the sample for the white balance tool is constantly 5x5 no matter what scale you are using in "Show Loupe". All Show Loupe does is show more or fewer pixels.
I think that 5x5 is a really small sample size and it would be better to be larger and best of all if you could vary its size.
Brad