Running MAX 2014 with FP 4.4.1. When I send a job to the farm of 12 computers, 8 of the computers render a single frame and then hang. If I unassign and reassign them to the job, those same 8 render one more frame and then hang again. I'm using your favorite, Mental Ray -

.
Network rendering has been surprisingly stable for a long time, but I did notice that the 4 computers rendering without hanging were older and running Backburner 2014. The other 8 computers are newer and some have up to MAX 2015 installed w Backburner 2015, a few others are up to MAX 2016 with Backburner 2016. This scene has a few FP objects with some sporadic grasses and trees, and then one heavy FP object with a medium grass cover. All FP geometry are mr_proxy objects. Initially, when I tried a wild grass variety from the library, this single frame thing happened. I then removed those wild grass objects, made something similar in Onyx and brought that in as an mr_proxy and with simple translucent A&D material and it worked great for probably 2 weeks of solid rendering and revisions. Then yesterday out of the blue, this started happening again where 8 computers won't render after one frame without the remove/assign exercise.
Troubleshooting I've tried:
1. updated every computer to FP 5.02 beta. Did not fix it.
2. uninstalled later versions of Backburner and installed 2014. Did not fix it.
3. put all nodes back to FP 4.4.1 and BB 2014 and no fix, but now broke the other 4 so now all 12 exhibit the same behavior.

4. rendering with "conserve memory" checked, no fix.
5. turning off heavy grass object fixes it, but I kinda need that grass.
Checking the logs, after the succesful frame finishes and the next frame starts I get a bunch of these errors until I deactivate the job:
"2015/12/30 23:44:04 DBG 3dsmax adapter : MENTAL RAY LOG: ***WARNING*** PHEN 0.13 6051 MB error: user data zero size."
with varying PHEN "X.XX" and "XXXX" MB errors.
Any tips on what to try next? Thanks so much for your time!!!