300 polys for a object? Are we suddenly back at the mid '90?
A modern graphic card can render 5-20 millions polygons without braking a sweat. A bit more if they are organized into tri-strips and fans, a bit less with quads and higher order polygons. And it's been this way for years. The limit is the number of draw calls (batches), you can have one mesh with 100 000 polygons but you can't have 1 000 meshes with 100 polygons (unless you merge them on the cpu before sending them to the gpu or use some trick like hardware instancing). Even if a model has 10 000 triangles the graphic card will only render 5-8k due to backface culling.
In the worst case scenario you will have 200-300 independent meshes (and that's waaaaay to many anyway), having them at 300 polygons each you get a grand total of 60-90k but only 40-60k (or less) will get send to the gpu thanks to backface culling. Even an Intel GMA can render 100k polygons at 60fps.
There are only two cases where polygon count matters - intensive animations or a complex vertex shader. I think FO uses the fix function pipeline anyway so the vertex shader is out, weapons are rigid by nature so there's no per vertex animations there, only characters are left so unless you put +100 bones in them your safe.
I can't help myself asking what is the target hardware for models with 100-300 polys?
And I can't help myself even more asking why not use a simple LOD mechanism? There'd be decent 5k models for close-ups and cheap 100-200 polys from far away.
