C++ Shared Library Performance

I am using the Matlab compiler to produce a C++ shared library on windows. The parent C++ program will iteratively call back and forth between Fortran and the Matlab generated shared library. When I run the m files from the Matlab GUI I notice that my CPU spins at about 30% and calculations are reasonably fast. When I use the shared library through C++ the shared library the performance is terrible. The Fortran code will spin my processor at 50%, then the matlab code will use 3% of my CPU. If I comment the Matlab calls out and time the code it is 60 times faster than it is with the calls to the Matlab shared library. How can I improve the performance? Do I need to add some options to my mclIntializeAppplication call? There are a lot of options I can use that appear to be undocumented. For example, the mclIntializeAppplication help page lists -fork as a valid option but I can't find anything that tells me what this does. Any help is or insight is greatly appreciated. Thanks.