tcp_keepalives_idle ignored

Hello all,

PG:  8.2.4

We have a very busy setup using multiple clusters, slony, etc.  My problem
relates to the number of postgres procs increasing, and not decreasing
when idle.  I eventually end up with thousands of idle processes listening
on /tmp/.s.PGSQL.5432 and not quitting (eventually bumping into
max_connections).

I have tried setting tcp_keepalives_idle = 120 (eg), then restarting PG,
but SHOW ALL; shows tcp_keepalives_idle=0 (ignoring my setting).

I have even tried setting setting /proc/sys/net/ipv4/tcp_keepalive_time
directly, but the idle procs just keep increasing in number.

tcp_keepalives_interval and tcp_keepalives_count I have left on default.
After a few hours worth of running, theres a few thousand idle postgres
procs, and they're all idle...

Any idea what's going on here?

Thanks
Henry