Marius,
My friend, I updated Startampr Wiki on Monday. As of version 2.3, my script reads that an IP on tunl0 is *required*.
At least on Linux, the outgoing IP on WAN takes priority, since it's the only one with a "Gateway IP" entry in /etc/interfaces. Routers can't "guess" which outbound to use. Once you coded -i into ampr-ripd, my discoveries disappeared, just like my pings using tunl0. Adding an IP has never had an effect, unless I wanted to test ping using tunl0. Never seen a "failure," as the IP is optional to the OS. Your coding on -i makes it required to use -L.
http://wiki.ampr.org/wiki/Startampr#Script
I also contacted Tom off-thread about it, as it seems he may be trying to run his machine as a desktop. If that's the case, his issue is that he has to configure the user applications to use tunl0, or simply virtualize the AMPR desktop and connect it's virtual-NIC to eth2.
73,
- Lynwood KB3VWG
But it determines the source address of the machine local outgoing traffic. It seems no one is bothered to correct this, even if it is classified as an error and dropped by the gateway.