Reduce CPU time, instruction cost, context switching, scheduler overhead, lock contention, and unnecessary computation while preserving correctness and end-to-end performance. Use when profiling proves CPU is a dominant constrained resource.
