When I started working on my engine I tackled things from an OO angle. Every class was clean and separate allowing for great orthogonality (sp?) between game objects. I can just throw together a bunch of textured quads, double sided quads, tile sheets or meshes in any order and things just work. Unfortunately this power comes at the price of performance. My engine definitely doesn't maximize the hardware. At GDC this year I had a little bit of validation on doing things this way when i went to the Tim Sweeny talk on Unreal Engine 3 and they do the same thing where they sacrifice performance to acheive orthogonality. I'm not even going to pretend my engine stands up to the readme.txt file of Unreal but it was still nice to hear I was on the right track.

As I'm getting closer to the finish line in my current project I'm starting to look for places to get the perf a little tighter. My renderer is definitely a great candidate. I have calls being made in each object that could be called only once per frame if I went about things a little different. The interesting thing to see is if I'll still be able to keep the level of orthogonality high. I have a goal of 50% perf improvement in the renderer which looks very achievable at first glance. Wish me luck.