Optimised RendererGroup invalidation, only the combined mesh holding the block is switched back to separate renderers Editor profiling, walking around the same big base: slowest LOD switch ~2.5ms -> ~1ms