Pavel Borisov explains the importance of vacuuming your tables:
By default, all table data in Postgres are physically stored using the “heap” method. So every database is a set of 1Gb files (”segments”) and each file is logically split into 8Kb pages. Actual table rows are put into any page with enough free space.
When the row data is updated, a new version of a whole row is constructed and written (to any free space). The old one remains because, at the time of the update, the transaction is not completed and can be rolled back in the future. When the transaction is completed we’ll have two or several versions of the same row in the table. Cleaning old ones is by an asynchronous process called vacuum (and autovacuum).
Read on for more information about how this works and what you can (or should) do to help it along.