You can use map-reduce on your search by partitioning up the starting keyspace alphabetically (26/num nodes), although with such an even partitioning the search will probably be unbalanced. But with some simple metrics on the starting letter frequency of words you could define a more balanced partitioning.
