preventing transaction wraparound

Lately I have been paranoid about the possibility of transaction wrap around
failure due to a potential orphaned toast table. I have yet to prove that I
have such an object in my database.. but I am running Postgres 8.3 with
auto_vacuum enabled and am doing nightly manual vacuums as well and cannot
explain the results of this query. Any assistance is greatly appreciated.

Yesterday I ran:

production=# select datname, age(datfrozenxid) from pg_database;
datname | age
------------+-----------
template1 | 100260769
template0 | 35997820
postgres | 100319291
stage | 100263734
production | 100319291

and today after the nightly vacuum ran I got this:

production=# select datname, age(datfrozenxid) from pg_database;
datname | age
------------+-----------
template1 | 100677381
template0 | 37594611
postgres | 100738854
stage | 100680248
production | 100738770

Am I just counting down to 2,000,000,000 and the postgresapocolypse? Is
there a way for me to determine what the actual transaction threshold is
going to be? I've read the postgresql docs and greg smiths section in high
performance and have to admit i am having difficulty understanding how this
number is not retreating after a database manual vacuum.

thanks,
Mike