On Hash-Based Work Distribution Methods for Parallel Best-First Search