Why is a multithreaded mex function slower when I used a parallel pool?

When I run a multithreaded mex function after having used a parallel pool, my function runs a lot slower (10-20x slowdown). Also, the more threads, the slower the function becomes.

Shutting down the pool does not help.

Rebooting Matlab restores the original speed (and scalability) of the mex Function.

[Walter Roberson]

Have you closed the parallel pool by the time you are running the mex, or is it still open?

[Walter Roberson]

Potentially related question: http://uk.mathworks.com/matlabcentral/answers/257315-can-i-speed-up-big-matrix-operation-with-parallel-computing-toolbox

[Bart Vermeulen]

Even after shutting down the parallel pool, the mex file runs much slower

[Walter Roberson]

I have a vague memory that someone else might have reported something like this a few years ago, but I am not sure I could find the posting. This is something I would recommend creating a support case for.