I have my big database replicate all tables and sequences via slony1 
(1.0.5).  I also take a nightly pg_dump of it for archival and disaster 
recovery situations.  Today I switched from dumping the origin to 
dumping the replica, but the dump file size was *way* smaller:

-rw-r--r--  1 root  wheel  2904749683 Mar  2 04:37 mm.02-Mar-2005.dump
-rw-r--r--  1 root  wheel  2916941855 Mar  3 04:38 mm.03-Mar-2005.dump
-rw-r--r--  1 root  wheel  2359870138 Mar  4 11:51 mm.04-Mar-2005.dump

the March 2 and 3 dumps were from the origin, and march 4 from the 
replica.  This is a "-Fc" format pg_dump of the entire database.

And yes, I'm still dumping the schema from the origin also :-)

The only other change was that the pg_dump client was upgraded from 
7.4.6 to 7.4.7.

The DB only really grows -- it only ever shrinks when we do our monthly 
cleanup of ancient data, which I haven't done yet this month...

What would make the DB dumps so much different in size short of missing 
data?  Could the ordering of the tuples in the database pages alter the 
compression significantly like this?