Are there any logs you can share?

Just as a note if the DUT has lost network connectivity it will reboot every 3 mins,  you can remove the file /mnt/stateful_partition/.labmachine to stop this reboot.

I fear with your unusual networking you might be triggering this network recovery loop, ggrundler can help with how this is triggered as he implemented it.

That does seem to be the reboot cadence that I am getting. 

(On the DUT) rm /mnt/stateful_partition/.labmachine worked. 

When the boot loop was happening, it would happen even if I take the DUT off moblab network. 

Yes the boot loop has nothing to do with moblab, test images that are deemed to be in the lab try to recover their usb network connection if it drops out, the last step in that recovery is to reboot the device, there are logs in /var/log/messages as to what other steps it takes.

Sadly all of this is necessary because we are not able to get a truly stable USB Ethernet connection when there are regular reboots.

Ah yes ... our friend recover_duts,

https://chromium.googlesource.com/chromiumos/platform/crostestutils/+/master/recover_duts/README

Should see something like this in syslog,

"All ethernet recovery methods have failed. Rebooting."

Sounds like the mystery is solved here.

Just want to add one more observation, once /mnt/stateful_partition/.labmachine exists and DUT is in boot loop, "repair" from moblab get stuck in the middle of the process. 

Are there logs you can share for the failure beyond 'reboot loop'?  Lets either attach those here and re-open after changing the summary or file a new bug to address that.