Did you read the article? In it he benchmarks Silk (accel) vs. Silk (no accel). He's proven to himself that the acceleration system doesn't work, using his own testing methodology, which, given that he produced the results, he should be biased into believing.
Nothing is keeping him from turning it on later in the future, when it's objectively proven to be faster. But he keeps it on now, basically because of his optimism in Amazon.
Nothing is keeping him from turning it on later in the future, when it's objectively proven to be faster. But he keeps it on now, basically because of his optimism in Amazon.