It depends a lot on what it is about. It for example isn´t the worst idea to have a complete level in one mesh with different materials, where you push all vertex data at once to the graphics card and then always setting the materials and drawing with the indices for that materials triangles. The fewer materials the faster it is of course, but it helps to reduce some of the CPU overhead one would have without sharing big chunks of vertex data.
Another example could be some kind of tool simulation, where you only have like 5 objects of which two or so have to be extremely detailed. But there it is of course the question if it in the end makes a visible difference between 60k and 200k polygons.
