As I understand it there must be matches for every field on the list to be a dup. So the more granular you get, the less chance of it finding a match. That's why addresses probably aren't a good criteria unless they're "normalized" both in your list and the values already in the system.