What's wrong with just using normal maps and displacement maps?

You start adding so much detail to a model and you're going to bog down, not to mention have all kinds of problems with bending the topology a hi res model necessitates.

That's why the pros, the people who make the movies and such, use very LO-res, UN-detailed models with very tight topology, in conjunction with normal maps and displacement maps and subdivision surfaces.

They would never finish anything using models with 2,000,000 polygons.