Of course people don't say it, but there are many cases where reported algorithmic improvements are attributable to poor baseline tuning or shoddy statistical treatment. Tao is exhibiting a lot more epistemic humility than most researchers who probably have stronger incentives to market their work and publish.