Wait, rereading the original queries I seem to have misunderstood something.
The individual parts of the partitioned tables are being accessed in timestamp
order. So what's missing is some way for the optimizer to know that the
resulting append results will still be in order. If it knew that all the
constraints were mutually exclusive and covered ascending ranges then it could
avoid doing the extra sort. Hm...
--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com