I was the one on your system yesterday. What I found out while on your system:
- The diskless root filesystem is mounted using the UDP version of NFS. I tried to change the initrd to mount it using the TCP version, but although on my system the change worked, on your system it still mounted with UDP. I don't understand why.
- The UDP packets from the MD that cause the NFS client to fail with "server not responding" are bigger than the MTU, thus they're fragmented. The problem on your system is that the last fragment wasn't getting to the Core at all, so the packet reconstruction failed and as a consequence the packets got dropped as invalid.
I left your diskless system with a shell prompt inside the initrd. You can press Control+D to exit it and continue, but chances are you'll get back to "NFS server not responding". You can also type "cat /proc/mounts" to see how the filesystems are mounted. One of the lines is longer and looks like this: "192.168.80.1:/usr/pluto/diskless/192.168.80.3 /mnt nfs rw,v3,rsize=32768,wsize=32768,hard,tcp,nolock,addr=192.168.80.1". The bold part is important to be there. Also, now I notice, those rsize and wsize values of 32768 may have something to do with it. All the other filesystems use sizes of 8192.