Are there performance caveats on index access when enforcing a pk with
a non-unique index?

I see that a query that specifies a single pk value against such a
table gets an INDEX RANGE SCAN rather than an INDEX UNIQUE SCAN. But
I'm wondering if it is actually more expensive in this case..?

I do not have test runs to support my assertion but from the few times
I played with the feature I did not see any performance difference so I
believe that for most non-unique indexes where few additional columns
not part of the PK are included that only a very minor performance
difference would exist. For non-unique indexes where no extra columns
exist in the index the practical difference should be zero.

IMHO -- Mark D Powell --