WUProp (nci) client using 48% CPU in Linux kernel 3.11.0

Observation:
after upgrading to kernel 3.11.0 I observed the Data Collection version 4 4.11 continuously using 48% CPU on 2 cores of a 6 core CPU and 50% of 1 core on a 4 core CPU.

It does not behave like this on the 2 core CPUs

I have never seen this before, so I assume that it has something to do with the changes in linux kernel 3.11.

Suspend and resume returns to the same state.

Detach & re-attach returns to the same state.

It is consistently not (nci) any more with Kernel 3.11

removing WUProp from these machines immediately

I have almost the same problem on only one computer.
Unlike AMDave is around 44% with a quad core processor. No matter whether other projects are running or WUProp idle.

It is a athlon 5350 (kabini) with 4 GB RAM, debian testing, kernel 3.14 1-amd64.

I suspect that there is some as the system. For example, the IGP supporting the proprietary amd graphic driver and app sdk, in most opencl wu's a similar cpu load is generated, in contrast to other computers.
Just what should have WUProp to do with it (I am almost sure that it somehow is related to the. But why?)?

The real point is that precisely the same software version of another system (AMD Kaveri) this effect does not show.

Oh, and one of two reported wu's has a strange way this CPU load is not:
21,602.24 19.04 14:00
21,535.27 6,760.51 14:00


edit: If I pause work and the directory /opt/AMDAPP rename (so that it is disabled) and start again the nci app back to normal (without cpu load).
I just do not understand the context; WUProp does not use OpenCL?