I am using QNX 6.4 SMP enabled kernel on 64 bit Xeon processor with 8 cores. 

To test the system, I have written a matrix multiplication program for 500X500 matrix size. I have created seven threads
 to do the multiplication that means each thread will have around 72 rows to work with.

Without setting thread affinity to a particular processor it is working fine, but when I set thread affinity of each 
thread on to each core, Time taken to finish the computation increase drastically. The profiling data shows that threads
 are in ready state but not running. No other proocess is running. It is only smp-procnto which keeps the core busy.

All my threads are with priority 255 and scheduling FIFO

Can someone please explain the behavior. Does this erratic  behavior has anything to do with a 32 bit operating system 
on a 64 bit processor.