I wonder whether some systems implement grep internally using hashes, i.e., the first time you grep a file it gets hashed and the system stores the hash in memory for subsequent greps. In that case, my fellow Anonymous Monk's original algorithm would have been as fast as a hash implementation.