‘Good to repeat’: Making random access near-optimal with repeated contentions