I learned something new today which on hindsight should be obvious: non-clustered indexes can be significantly faster for index scans than clustered indexes, because they can take up far fewer pages on disk, especially for tables with many fields. Read VRohr’s blog for details.