Heh, it's so incredibly fast it's not even funny
Still stuff I can do... in particular, using vertex buffers where supported. Right now, each end octree node has a compiled display list on it that we execute to render that node, and replacing that with a vertex buffer in video RAM might help a bit.