The specific implementation of your greylisting is not of concern to people running big MTAs. There are many implementations of greylisting and spam filtering in general. Pick another one that works for the kinds of email from the kinds of people you expect to receive.
Greylisting works on tuples, one part of the tuple is the originating IP address. If you have a bunch of servers with different public facing IP addresses retrying the same email, then that makes it a much greater workload to finally offload the email to the recipient server.
My greylist implementation had a bit of trouble with this initially, but the volume was enough that it learned to trust the originating IP address + sender domain alone. A small volume of email makes automatic learning this information harder though, so manual whitelisting might be required. James