Just being able to rerun the benchmarks (or other data analysis for non-CS papers) would be an improvement on the current state of affairs, where people often don't publish code nor data.