natd[PID]: failed to write packet back (No route to host)

Hello,

I have a client version of 10.6.4, acting as a gateway server with IPFW2, NATD, DHCP3, DNS services running. I've had this setup working reliably for many years now (with all different versions of Mac OS X Client). My server has two ethernet interfaces:

en3: for the LAN (192.168.0.x)
en4: for the WAN (static IP from ISP)

[for the record en0 is the ethernet jack - which is fried, en1 is the airport which I don't use, en2 is lost in the ether somewhere (pun intended) from when I had a different USB->Ethernet adapter]

It seems that I have a natd problem. I constantly get this error in the /var/log/system.log:

natd[PID]: failed to write packet back (No route to host)
natd[PID]: failed to write packet back (No route to host)
natd[PID]: failed to write packet back (No route to host)
natd[PID]: failed to write packet back (No route to host)
natd[PID]: failed to write packet back (No route to host)
natd[PID]: failed to write packet back (No route to host)

And neither the gateway, nor the clients that it serve can connect to the internet.

When I run the Network Utility.app and view the en4 interface, I can see that there are many more (magnitudes more) received packets than sent packets. Traffic on the LAN (and on the en3 interface) is fine (as long as the traffic is LAN only).

I have solved the problem with the following script:

#!/bin/sh
killall natd
sleep 15
/usr/sbin/natd -f /etc/natd.conf
exit 0

But I have to manually run this script after every restart. I've tried to load it at startup automatically, but that doesn't work (maybe because services load in parallel, and natd is starting too soon).

Anyone know how I can fix this so that it doesn't happen anymore? This situation makes off-site restarting of the server impossible (as infrequent as that may be).

Thank you in advance.