I was hoping that Dave Etchells would get time to take the requested shots for some better analysis, but it's gone quiet.
As far as I understand, there will be much more noise in the auxiliary subframe - this likely explains Northrup's finding that using the Dual Pixel raw features with a high ISO image can seriously increase noise (because it's likely having to use "half" the data if you do a large swing with the dual pixel feature).
However, the assumption is that for the top stop of highlights, the extra noise should be acceptable (good signal to noise ratio) and thus you could replace a stop of blown highlights from the main subframe with a pushed version from the auxiliary. As far as I know, no one's done that directly from the raw data and analysed the DR yet though - so it looks very promising, but there's no absolute proof it'll be a real increase in DR.
Being a pedant: those images are slightly misleading, in the sense that DR is defined as the ratio between the largest and smallest signal that can be recorded (with an acceptable level of noise). Different sensors have different well capacities, but if you shot the same scene with each camera, and put the brightest highlight at each sensor's saturation point (just below clipping) then the DR would be defined as how many stops down into shadow detail you can go before noise is objectionable.
Putting it another way: different sensors have different 100% points, but there's always a 100% point. The crucial bit is what percentage still gives a sufficiently clean result (i.e. it's always about the darks 
)
[I see the DPReview chooses a tone as "middle gray" then measures outwards either way - which explains the images. Not a "classical" way of defining DR, but not invalid]
I think that is more relevant than hitting the buffer.
