Demonstrated in https://forum.pjrc.com/threads/54711-Teensy-4-0-First-Beta-Test?p=212280&viewfull=1#post212280

Does this have possible utility worthy of supporting?

Did a compare of overhead factor -10 to -13 and change to -11 tracks better.

Did a loop of 20 passes of 5 inline calls and divided the CycCnt /100 - using offset of -10 tracks better than -12.
>> uint32_t begin = ARM_DWT_CYCCNT-10; // Overhead Factor for execution

Calling delayCycles() cusps on groups of 4 cycles because of while().
returns in 6 cycles up to 13
then 15 cycles up to 21
then 23 and counting cycles starting at 22 in groups of 4.
* integer math rounded down - put table in the forum post.

Tested at 600 and 300 MHz and looked the same - of course that is twice as long :)