The way PXE boot works with rc.initdiskless is the client does some work to assemble its own filesystem. I'd prefer a PXE boot layout where rc.initdiskless does as little as possible. In other words, PXE client A does not have access to the configs specific to PXE client B and vice-versa. This means doing most of the work on the host side. Ideally, each PXE client would see a different file system via TFTP and NFS.