"David Rowley" <dgrowley(at)gmail(dot)com> writes:
> I assume workmem, effective_cache_size and random_page_cost are all the same
> in the 2 postgresql.conf?
Indeed, work_mem is probably the problem. The critical difference
between the two plans seems to be that the first one is using a
"hashed subplan" and the second one isn't. Assuming the same datatypes
in both databases, the only reason not to use a hashed subplan is if
the hashtable is estimated not to fit in work_mem.
regards, tom lane