Difference between CPU and GPU time in the profiler

I’m trying to use the opencl profiler on my program but I get strange result for the CPU time of the kernel execution.
While the GPU time is always approximately the same (around 5 350ms), the CPU time is often around 6 000ms but somtimes takes really big values (93 000ms) even if the GPU time does not change.
What do you think it could be due to?
The program is kind of big and multi threaded and using OpenGL on another thread. Could the time difference between the GPU and the CPU come from the use of OpenGL? I guess that the use of OpenGL in the meantime as OpenCL should slow the execution of OpenCL down but this slowdown shouldn’t be counted in the GPU time?
Has anybody had this kind of problem before?