I agree that the problem as stated is fairly interesting. However, I wonder of maybe we have an X/Y situation here: rather than figuring out how to find the bottleneck(s) in your current algorithm, perhaps we could find a better algorithm for your underlying problem. Which is... ?