That is the big question, Unigine uses an unusual amount of tesselation at the expense of other things. Games will most likely not use it to such an extent and there is still the question of how nvidia will be able to utilise their tesselation units when the shaders they are linked to are utilised for other functions.
even with these advantages, nvidia had to run the benchmark with no antialias and 1x anisotropic filtering. What that does is allow the polymorph engine to concentrate fully on tesselating, showing it at its maximum strength.
just typical nVidia smoke and mirrors, if they had showed it against a 5970 it would lose by a lot.
With the PolyMorph Engine, GF100 implements the world’s first scalable geometry pipeline with up to 16 individual tessellation engines on a single GPU. These engines deliver revolutionary performance in DirectX 11’s most important new graphics feature, GPU-accelerated tessellation. By allowing more detailed geometry to be packed into a scene, tessellation lets developers create more complex environments with beautiful visual clarity. Blocky edges are smoothed out, allowing game characters to be rendered with cinema-quality detail. And since tessellation is only applied to necessary areas, game performance is not compromised.
To me, it means with heavy tessellation, it can be comformed to handle a ton of tessellation at the expense of normal processing power. This will mean that dev's will have to balance it's use carefully, it also means that with a lot of tessellation, it "can" perform a lot better than the 5000 series.
What that does not say is how well it performs normal tasks. It might not compromise the performance the card can normally handle, but it doesn't say it will perform up to the standards of any specific card.
The concept is cool, and may be the wave of the future. How well it works today is still up in the air until we see an array of benchmarks. You can bet the cards performance is going to have peaks and valleys, much more so than anything we've seen so far.