I am fairly confident that is a typo. The paper says "Our analysis reveals that RNG topologies are 9–45% cheaper than fat trees with equivalent oversubscription

I am fairly confident that is a typo. The paper says "Our analysis reveals that RNG topologies are 9–45% cheaper than fat trees with equivalent oversubscription ratio." The blogger probably left off the % sign.