The fact that this provides the benefits of both approaches, namely that it is (branch) predictable and amortises memory latency while obtaining the probabilistic benefits of double hashing.