He's not saying anything about punishing anyone, just about stopping to reward quantity over quality. That's nicely compatible with rewarding replicable research, as replicability is a sign of quality.