We did not fully figure it out. The client was doing several hundred network devices at one time. Fix was to split them into several schedules, but in reality this was not how the client would end up running schedules. 

Are you running scans on just network devices? If so, how many? 

It was definitely strange, they only thing I could think of would be QoS on the network and discovery timing out.

However, the client is now running larger scans by subnet and we haven't seen any issues.