What that is really referring to is applying high megapixels to a print or a standard output size.
You can make the dots of ink smaller, in which case noise becomes insignificant. But printers already have their limits and your optimum remains 260-300 dpi.
So instead you would "downsize" the image - combining 3 pixels of a 36 MP image into a single pixel for rendering in a normal print. This means that noise is a small amount of the result, and virtually disappears.
DxO uses a standard size print to evaluate all the different cameras with different sensor sizes. That print size is 30x20cm or approximately 12x8 inches.
You could pick an alternative size - 30x20 inches inf you wanted. That requires upsizing. That would mean that each pixel of a D3 needs to be converted to 2 pixels - and any noise would result in twice as much noise.
You would not be throwing anything away - just evaluating based on a common output. But if you want to compare images from two FX cameras, the idea is you should compare the same size print from both rather than a much larger print on a camera with more pixels.