10% less time for the same amount of work, since we are measuring the total response time of the workload in question and looking at median (p50), p90 and p99 numbers.
At it's heart they seem to be asking kind of a latency vs. throughput question. The answer is frustratingly "it depends" but for some real world Rails benchmarks, Shopify has published quite a few benchmarks including the ones in the linked article.