There actually appears to be conflicting information on how DGO works from Canon. TDP has a quote from Canon's last digital event about it, that suggests to me the two image halfes are involved. Which also seemingly contradicts their other marketing. And of course, he only mentions that there are no temporal artifacts, but stay quiet about those from parallax.
"How does it do that? Each pixel on the sensor is split into two different diodes. Those diodes are always taking two frames of the exact same image. Now for Dual Pixel autofocus, it's using those two frames for phase detection. However, the Canon engineers realized that they can use those two different frames and value them at different gains in order to expand the dynamic range. And so what is happening here is that off of diode A you have one image that is low gain and low noise. Off of diode B you're getting a frame that is of higher gain, but it satisfies the pixel's need for saturation. So, these two separate frames that are of the same image of the exact same point in time but at two separate gains are combined and then dumped out of the sensor. Because these frames are of the exact same point in time and of the exact same image, there is absolutely zero temporal artifacting happening in this" (
Source)
Do you have a better source for how Canon has pulled off DGO? I would be very interested, as the marketing I've seen seems somewhat lacking as far as proper explanations go.