On the Benefit of Sub-optimality within the Divide-and-Evolve Scheme