You have it back to front - tesselation saves resources. If you want a 100k detail poly character without tesselation, you need 100k vertices in RAM. However with tesselation, you can have a 25k model in RAM and add the extra vertices. You can also choose when to add vertices, allowing simpler geometry at a distance for LOD savings, or even if there's a lot on screen, reducing the per-object complexity as demans increase.
These platforms have enough vertex power to render a polygon for every pixel, meaning 'perfect' geometry detail, but we haven't an efficient means of managing the workload. Tesselation is a step towards more optimal polygons, where the ultimate goal would be one triangle per pixel and no undrawn triangles.