Removing the "min of averages" model and trying just the minimum result to see if it is acceptable. It'll at least be faster.