With an i5-4690k , more than 3 instances of the miner with 1060s, the hashrate of every instance falls dramatically

When the miner crases, the error relates to libstdc++-6.dll


Can you put a parameter on miner to set -cl-local-work?

Perhaps the solution another miner developer adopted may be useful to you

He made a opencl miner that eat all cpu available, and the more gpus working, less performance for each one

There is an issue with NVIDIA GPUs leading to very high CPU usage. The reason is improper implementation of OpenCL by NVIDIA. When CPU thread waits for results from GPU, it does not stop, it spins in loop eating CPU resources for nothing.  
 There was impemented a workaround on this issue: before reading results from GPU current thread sleeps during small calculated time.

He resolved it and the miner is more stable and a lot faster.