It's possible that if a managed domain operation that sends a request to another process in the domain is cancelled immediately after initiating sending of that request that the cancellation request will be received before the original request. If that occurs, the cancellation request will fail (as there is no ActiveOperation to cancel) and then the original request can hang waiting for a tx commit/rollback that will never arrive.

See http://brontes.lab.eng.brq.redhat.com/viewLog.html?buildId=15650&tab=buildResultsDiv&buildTypeId=WF_MasterLinux#testNameId2482043836298342768 for a case where this occurred.