Resolution:

1).       Figure out which thread is using the most CPU (Use top or ps command on UNIX).

2).       Generate a thread dump several times. We would like to see what the RVRD is doing.

3).       If the high CPU thread is similar to the following thread trace:

#0  0x000000000048a784 in _tibrvHash_Get ()
#1  0x000000000041ccd5 in isSubscriptionInSessionList ()
#2  0x000000000041d091 in removeSubjectFromDefunctSessions ()
#3  0x00000000004183e8 in continueExtendedOperation ()
#4  0x000000000041c9c0 in fireTimer ()
#5  0x00000000004a508a in _tibrvQueue_DispatchEx ()
#6  0x000000000049b46e in _tibrvEvm_TimerThread ()
#7  0x00000000004acd4e in _tibrvPool_ThreadHook ()
#8  0x00000000004b3cfe in _tibrvThread_HookWrapper ()
#9  0x00000036dd007851 in start_thread () from /lib64/libpthread.so.0
#10 0x00000036dcce811d in clone () from /lib64/libc.so.6

Run tibrvlisten to listener “_RV.>" . The output should cover the period before the problem occurs. If you 
notice “SESSION.STOP

“protocol messages", you may see :

RV-3201: Improved routing daemon processing of release 5.x client shutdowns when a routing daemon is tracking many subscriptions.

This backlog issue only exists with V5.5 or earlier version clients and does not exist with any client version newer than V5.5. 

This issue was first addressed in RV 8.4.1 hotfix2.  Otherwise,  send  the thread trace and tibrvlisten output to Tibco Support for review.