There is another problem that I noticed while looking at this case. The analyze took close to 500sec on a fairly good hardware (40GB RAM, 10K rpm disks on RAID10) because many large child tables were scanned at once.
Just a small correction to get a good benchmark
The ~500 sec analyze time was on a test VM running on a i5 2.4 Ghz with 2 dedicate cores, 4 GB of RAM and stored on a notebook HDD. The partitioned data was about 80GB.
On our production environment (described by Pavan) it took ~90 second for ~55GB of data in 8 child/partition tables (we stopped the import of partitioned data when we discovered this issue - we COPYed and TRUNCATEd partitions to speed up upgrade procedure from 8.4 to 9.1 by pg_dump/pg_restore).
