5 cars = 5 images (or 25 if they had 4 distinct wheels each) ... the amount of poly isn't the problem ... the amount of distinct objects that need to be transformed thought is ...
Show me a 2D game where you need less than 100-200 images ...
and thats the point where missing TnL will start to become a problem at least if you try to rotate and scale them in realtime all the time.
To work great on low end cards, a more batching oriented system that reduces the amount of transformation state changes would be very suitable. On modern GPUs that driver that once was created for DX7 didn't do much usefull ... but on a stone age pre 3D cards (intel before GMA900 and the like) it would reduce the amount of matrix multiplications that have to be done by the CPU.
Show me a 2D game where you need less than 100-200 images ...
and thats the point where missing TnL will start to become a problem at least if you try to rotate and scale them in realtime all the time.
To work great on low end cards, a more batching oriented system that reduces the amount of transformation state changes would be very suitable. On modern GPUs that driver that once was created for DX7 didn't do much usefull ... but on a stone age pre 3D cards (intel before GMA900 and the like) it would reduce the amount of matrix multiplications that have to be done by the CPU.