A µC has a lot of signal processing hardware in it (timers, ADC...) that is already used by the framework (for timing). If a software unit uses part of that hardware that is also used by other units then it's useless to test both units independently, because they probably can not cooperate. Or it's only the computation time of one unit that prevents another unit from working properly. µC coders face such concurrency problems every day, and no single unit test can help them out 