This paper describes a performance evaluation of different implementation alternatives of a parallel C++ program for telecommunication applications. The implementation alternatives use different combinations of Solaris threads and Unix processes. A Sun multiprocessor with eight processors was used in the performance evaluation. The measurements show that the multiprocessor system favors implementations with many Unix processes, containing a small number of threads each, compared to implementations with one process containing a large number of threads. It also turns out that the performance can be increased by binding threads to processors, and that the overhead grows significantly when the number of processors increases.