@EdgarBrown The resistive nature of MOSFETs leads to power losses quadratic with the current, while the fixed voltage drop of BJTs leads to linear power dissipation. Then, FETs dissipate more power than bipolars, at least asymptotically. Are the constants involved the key, or I am being wrong elsewhere ?

@SolveEtCoagula07, yes it is exactly what is needed, thank you !

Sorry, I trusted the website (it says "product data sheet"... ), but didn't downloaded it again. Nonetheless, I searched in the datasheet, not in this brochure.

Here

Sorry, I misexpressed myself: I've solved the problem, it came from the Wait loop, and not from anything related to timer1. In this Wait loop, I looped until T > 1000, instead of looping until T >= 1000. Changing that, all works correctly, and I've only an error of 0.01 %, in average. So, the correct setting is OCR1A = 15999. Thank you very much for your help

Finally I found the problem: now, the error is about 0.1 µs for each delayed microsecond.