delayMicroseconds uncorrect, about 50 ticks too much

I measured the following delays in CPU ticks (80 MHz):

```
delayMicroseconds(1) : 130 ticks, 50 ticks too much
delayMicroseconds(2) : 202 ticks, 42 ticks too much
delayMicroseconds(3) : 286 ticks, 46 ticks too much
delayMicroseconds(4) : 370 ticks, 50 ticks too much
delayMicroseconds(5) : 442 ticks, 42 ticks too much
```

Couldn't a precise function be implemented?

I measured these values during a NMI interrupt, the behaviour is always the same and reproducable.