For the one that now works, yes I do use Journal based backup. In order to get the first full pass I had to use the memoryef switch. I could backup each of 2 higher level directories without the switch, but a full pass of the volume would fail with memory allocation errors until I used the switch.
The other system I am hesitant to make any changes on, it is our document imaging system and contains its own TSM instance that runs at a 4.1.4 level (the version of Content manager can't run any higher due to the API's it uses). The backup client is at 5.1.5 and my backup servers are at 5.2.7, so I do have the option of JournalBB, but if I can't get through the volume on one pass, the journal is never ready to use. And since that level of TSM has a memory leak, it is scheduled to reboot twice a week, meaning a regular incremental would be needed twice a week before the journal is ready to use.
Our migration of that system to a newer version (including the TSM instance) has been in the works for over a year. Until then, I'm may be screwed. This problem has only started occurring recently, I am opening a PMR today, but I don't hold out much hope unless there is some client or server setting that can remedy the situation. I've already set commtimeout and idletimeout to 2 hours, I may try even longer hoping I don't get other clients that get in an idlewait state and I hit my max client sessions.
If you have any other insights, I'd be happy to entertain them.