There's no need for perfect repeatability when statistical analysis is good enough. After all, even if you have control down to the iron, the randomness in external interrupts and the effect of temperature on the hardware will cause some unpredictability.