> Why would that be? I would think that scanning the index for IDObj to
> conclude the value does not exist should have the same cost for all three
> tables. It should never have to read the actual datapages, and the index
> size should be almost the same for these three tables.

+1 ... not understand why it should be something else