It looks like they're comparing their new "Blender" search server with a specific case when search volume was higher than average on their old server... while this is definitely faster, I don't think it's 3x faster on average.

>> "Following the launch of Blender, our 95th percentile latencies were reduced by 3x from 800ms to 250ms and CPU load on our front-end servers was cut in half."

It doesn't look like that 250ms latency is at their 95th percentile search volume. It looks more like the average dropped from 600ms to 250ms, which is still very impressive.