This is off topic so I'll be closing this thread but in general you want to structure your parallel data to perform substantial work on non contested memory to both be worth the overhead and actually...