When it comes to multithreaded applications, a number of elements play significan roles in how threads run. Using multicore CPUs is quite a prerequisite for such type of applications when you leave the basic level and move to using several threads and/or performing CPU-intensive tasks.
Going to the software side, first of all you must carefully exclude from your threaded routines all time consuming parts that can be postponed or moved to the main thread.
Second, you could try raising the execution time of threaded functions calling SetSleepPolicy (VAL_SLEEP_NONE); entering them; this usually has better results if you create a thread pool and specify CmtSetThreadPoolAttribute (ThreadPool, ATTR_TP_PROCESS_EVENTS_WHILE_WAITING, TRUE); for the thread pool.
Additionally, individual thread priority can be set if you use CmtScheduleThreadPoolFunctionAdv: see this help topic and related arguments in the online help. I would suggest trying the sleep policy first and playing woth the thread priority only in case you keep having errors.
A prerequisite of all this is that you are sure you don't have code problems in your application and the GPF is due only to CPU load. You should particularly debug all your resource-accessing calls in order to be sure that no concurrent access to shared resources, variables or memory blocks can be possible; using thread-safe variables and/or locks to prevent concurrent access by different threads to critical objects can help in this matter.
