It's actually based more on a "points" system, of which I'm trying to develop. If your models use the usual 2:1 polygon:vertex ratio, then 4000 to 5000 is the best area to focus toward. If you have a 1:1 ratio (an equal number of vertices to polygons), then you'd focus more into the 2500 to 3200-polygon range as both have about the same number of "points". You can see where the peak is by using my benchmark tester program and looking at the data.