Have you compared this solution with tools like Preprocess, Reducto, etc.. ? I'm curious about the performance gain you can achieve with your approach