how much orchestration is too much with Ollama

[localpatch78]

I think the bottleneck is somewhere else. I get why that seems right, but on dual 3090 the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.