I guess the high-res maximums are 262.14 ms / F_CPU in MH for _delay_ms() and 768 us / F_CPU in MHz for _dalay_us() regardless of the availability of __builtin_avr_delay_cycles(), but the way the _delay_ms() and _delay_us() descriptions are written this is not entirely clear. Am I correct?
If so, I'd suggest rewriting the descriptions for those functions to not re-use the term "maximal possible delay". Instead they could say something like "maximum possible high-resolution delay" and "maximum possible low-resolution delay". Britton