First off, I would (and there is a similar thought elsewhere in this thread) set up any "Non-OS" packages that you install in their own nest complete with the customizations you need. Once that is done I'd create RPMS of it all and then write those RPMS someplace safe. CDs will work or even (how quaint!) tapes.

Secondly, I would keep that anaconda.conf file that gets generated when you build the system from scratch and use that to set up a kickstart environment for your servers. This will pay off in simplifying the task of rebuilding your servers. If you further modify the kickstart with your RPMS that you made and integrate your RPMS onto your kickstart server you have further enhanced your recovery process.

Thirdly, set up some sort of CVS mechanism whereby you only modify configuration files system wide and keep them in a CVS repository. A deployment script could then be used to check your changes out of CVS and write them to your production system. Optionally you could write some sanity checking into your deployment scripts (I leave how as an intellectual excersize for you to solve).

Last thought: Will all the above you could write postinstall scripts to run during the kickstart build to perform any last moment tweaking, CVS checkouts or whatever to bring your system up to a known state. It is even possible for you to do a restore from tape or whatever from inside the post install scripts.

Linux and the Open Source Community provide lots of tools to get this job done and with Kickstart, well... Kickstart is your friend.

I just helped someone set up an environment where they keep their Kickstart servers turned off except when they need to do a rebuild. Then the boot up the KS, reboot the machine that needs rebuilding and away they go. In fact they will rebuild their machines while the standby machine takes over once in a while just to make sure their procedures work.