Well, I think you are misunderstanding something. Your premise is that Dual Pixel architecture means Canon is throwing away half the light of each pixel, right?
Well, from my understanding of the dual raw format, it is actually using the fact that both the sum (which you would get in a regular CR2) and individual value of one of the halves is given. The halfes only have half the capacity, so they are essentially one stop underexposed compared to the sum. This is used in DPRAW extraction to gain 1 stop of highlight DR, by merging the single half value into the sum, where that is clipped. That's just my understanding.
A better proof that combination of the two halfes occurs is probably Canon themselves. Quote:
"During AF detection, the two halves of each pixel -- the two photodiodes -- send separate signals, which are analyzed for focus information. Then, an instant later when an actual image or video frame is recorded, the two separate signals from each pixel are combined into one single one, for image capturing purposes. This greatly improves AF speed over the majority of the area on which you’re focusing." -
Source