Cluster(ing) to help joins?

Hi there.

I'm working with a DB under 6.3.2 that has a little over 2million rows of
data spread across 8-10 tables. The question I have is when I attempt to
join more than 3 tables I get a HUGE performence hit, even if the number
of rows returned by the first set of clauses is small.

If I perform part of the join minus a single table the performance is great.
So why is this? Can using clusters help in this kind of situation?

I did discover that building multi-key indexes did help the performance a bit
but it is still way too slow at times.

Oh...and can someone tell me what this error message means?

ERROR:  ExecInitIndexScan: both left and right op's are rel-vars

Thanks for any and all help...

-=pierre