I have experienced similar issues, and here are some fixes that worked for me:
1. Our nightly Windows server backup solution was backing up my NCM server during its scheduled switch/router configuration backup jobs. As a result, the backups were failing. Rescheduling the backup of the Orion server to take place much later in the day resolved that problem, and more Orion switch backups completed successfully.
2. I opened a case with SolarWinds for further troubleshooting. They requested confirming that Windows UAC was disabled. It was, so no impact there, but you might check it.
3. Further into that same SolarWinds case, they recommended reinstalling Job Engine v1, and that was helpful towards correcting more backups failures.
4. Next was manually reassigning templates to the specific Cisco devices that were failing backups. For a great product, this shouldn't be necessary, and sometimes backups successfully without templates being manually assigned, but I did the manual template assignments anyway. I didn't see a significant improvement after doing this, but it's easy to try.
5. Lower NCM's Simultaneous Downloads to 1 and rerun the job. This had positive impact, and eventually I increased simultaneous backups until it was at 6, which seemed to be a sweet spot on my network. Higher than that and I saw additional problems. The default was 10, if I recall correctly.
6. Changing the timeout for each connection backup session seemed to have the biggest positive impact. Apparently NCM takes a lot longer to download & recognize through fuzzy logic the contents of the configuration files, and I'd recently added a lot of new Cisco 4510's that were fully populated. Their larger configs took longer to download than the timeout settings allowed, so the backup sessions shut down before the configs were completely downloaded.
Good luck on resolving the issue! Please post the resolution once you find it.
Swift packets!
Rick S.