Based on rsiedl's description of the setup, it sounded like there was already a rational basis for segmenting the data into distinct tables, and the usage pattern for the data didn't seem to motivate full integration into a single huge table (rows fall into large independent sets, and queries never need to span across sets).

I'll confess that I'm no jedi when it comes to optimizing an RDB through careful crafting of table layout, query construction and indexing. This probably explains why I can imagine situations where the conditions on an oft-needed query might not lend themselves easily to effective indexing, and reducing the size of a full-table scan would be worthwhile.