What RBL lists are you using? What exactly is the problem you are having? Just that users are getting lots of spam or that the mail server is being overloaded? How much mail are you getting (per second?)

Configured correctly, Postfix running with a spam filter and possibly greylisting should help. Check out this pre-data anti-uce configuration for Postfix, which should weed out a lot of your spam - although replace the spamhaus rbls with zen.spamhaus.org.

Using multiple rbl lists will delay email and take up processing time, if you're not using a local caching nameserver then this can be a considerable delay. It's also important you do the rbl checking in the correct place - ie. local users don't need rbl checking, and cheap local checks (eg. invalid hostnames) should be done first.

Another tip not often documented is the use of fail2ban alongside Postfix. If you get it denying connections from ip addresses causing rbl failures, relaying attempts, etc. then it'll reduce the amount of connections.

If you're building a new machine to do the filtering then you'll want to train any bayes filters with existing data - have you got a way of doing this? If not, you might want to get the existing setup collecting spam and ham now.

After all of this, if you've still got questions then there's the #postfix channel on freenode. They're a friendly bunch provided you don't ask stupid questions.

I've written a little on this site about filtering with Postfix on Debian.

In Etch you have policyd-weight packaged, which is a good way of picking up the best of block lists, along side greylisting (with postgrey), this is a pretty effective first line defense against spam.

Of course without knowing what sort of spam they are suffering with it might not be appropriate.

Ultimately the IP address reputation based filtering is suffering because of the amount of spam now coming from Gmail, Hotmail, AOL, and other large email providers. If this continues one will have to either blacklist these providers, or use some sort of content based filtering.