How to organize multi-task implementations using dependency-aware waves for maximum parallelism - and why theoretical speedups rarely match reality.