I suspect something isn't happy about 2 tasks/card;

-instances_per_device 2 is missing in your linux version.

Some params are not working on Linux as i found on my testing last year.

I've never been sure that this parameter mattered; it felt like it was merely for logging. I was passing it for stock and it wasn't logging and I was passing it for the build TBar made, too, and I just checked and it wasn't showing up in the output there either.

I suppose I could go trawling through the code to see where it's omitted on Linux but since I haven't heard anything back from my last submitted patch my enthusiasm is low.

I checked 16.10, the release date was 13-Oct-2016, http://releases.ubuntu.com/
I've never had much confidence in brand new Ubuntu releases.
nVidia has released One non-Beta driver since then, http://www.nvidia.com/object/linux-amd64-display-archive.html
I'd give that OS version a few months before I tried running SETI on it.

I built a couple Apps from r3568 and they are both slower on my machine than the 2 r3566 Apps. The r3567 build a CA is the Best App on my machine at present. I haven't had any stalls with any of the Apps and NVIDIA X Server Settings has the GPU load bouncing between 96-100% running one instance.
r3568 SoG build, http://setiathome.berkeley.edu/result.php?resultid=5308398478
r3567 aka r3566 using Intel path, http://setiathome.berkeley.edu/result.php?resultid=5308398484

There seems to be a problem with the r3568 Apps and the reference_work_unit_r3215.wu. For some reason that WU will produce an Error in Standalone mode while none of the other WUs have that problem,

...
period_iterations_num=10
ERROR: OpenCL kernel/call 'clGetEventProfilingInfo' call failed (-7) in file ../../src/GPU_lock.cpp near line 546.
Waiting 30 sec before restart...

FWIW I've run that build for nearly 24hrs with single task per card and I'm not seeing any of them stall out nor am I seeing driver crashes.