I have a discrepancy in database size that I don't quite understand. I'm using postgres 8.1.3 on solaris 10. I have two databases on the same machine - the production database and a staging database. The production database is 9 GB, and growing fairly rapidly. Recently I did a dump and restore of the production database to the staging database. The recreated staging database was only 2 GB (approximately). There were no errors in the reload.
I compared the two databases in terms of number of records, and then actual table size. At first I thought that there was a problem with the connection pool in the application which in turn was keeping large temporary tables around. This was part of the problem but only accounted for less than 1 GB of the difference. The real tables are different sizes. The ones that seem to show the most difference are tables that have the most churn: much of the data is deleted and reinserted nightly.
The database server has autovacuuming turned on, although the application also vacuums on a schedule.
You might want to do a vacuum verbose on some of the suspect tables to see how many dead tuples there actually are.