Prior to seeing the pulse start, the timeout would tick by at 16/11ths of the speed it should because the first two loops - the first two loops aren't also incrementing width, which requires 5 cycles, so of course they'll run faster! This change adds 5 cycles of nops to each loop (3 instructions - 2 rjmp+0, 1 nop). This cost 6 instruction words of flash - however, the compiler missed a chance to share the set of pop instructions between the two paths to return; doing this saved 6 instruction words, getting us back the flash the timing fix took away.