People have been gaming ML benchmarks as long as there have been ML benchmarks. That's why it's better to see if other researchers are incorporating a technique into their actual models rather than 'is this paper the bold entry in a benchmark table'. But it takes longer.