I have not heard of such problems before. How is NNM determining these interfaces are down? Is it relying on SNMP traps, or is it using polling? Can you get a packet capture of all traffic between the NNM server and such a device when LMS is taking a configuration backup? If so, post the capture file.