I think the problem is that the indexes on the words are not usable forregular expression matching. If the words are folded to lower casewhen the index is built, then using an exact match (or even like) shouldbe much faster.