He generally explains that Canon has had, and still holds, a marked advantage on read noise and well size. On these the D3 is behind the much older 5D and 1Ds Mk II even.
Where the D3 falls off the chart (literally) is base sensitivity, expressed in signal at unity gain or effective “ISO per electron”. The D3 is very sensitive to light – with better signal enough to overcome worse noise, the final S/N ratio is improved.
It’s worth noting that the P&S in the comparison actually have much lower read noise, but that noise crushes the poor signal easily. Sort of the opposite of the D3.
The document ends on an image quality chart that essentially says that at base ISO, electron wells and pixels matter. But each of these sensors gets different base ISO (recall ISO at unity gain). By the time the Mk III gets to the D3’s base ISO it has matched the other’s S/N, and thereafter it’s trading resolution against S/N until it falls off the wagon at 3200.