Let's look at top surface. To have it 2 times wider horizontally than vertically in projection camera angle must be 30 deg down from horizon.
Ratio between the foreshortened diagonal of top surface and foreshortened height of the cube would be sin(30)/cos(30) if they were same length in 3d.
Since they don't have the same length in 3d (diagonal is sqrt(2) times longer than cube height) the final ratio in projection between cube height and diagonal comes out to be cos(30)/(sin(30)*sqrt(2)) so more than 19px and less than 20px (a little bit closer to 20px) for the foreshortened diagonal of 16px.