All fine and good but think for a second.

The video card is only going to render the points that you can see.....
But the PC's memory and the application has to be handling ALL POSSIBLE points,
from ALL POSSIBLE angles at all times...

Also, think how much more cpu-hungry this will make any "poly to poly" collision tests...

And why does the demo have so few "different" models? "We havent made many yet."
Lame excuse methinks. Who would release a next-gen game demo with low poly models?
Thats what thay are doing. It just looks like model-instancing to me...