True... I wasn't 100% on it myself but I was just a little ruffled to be kinda glazed-over being practically the first person to acknowledge that accurate timing even could be done, given accurately-timed code 
Just mostly ruffled that the topic transitioned from "it can't be done" to "here's how to do it" without even recognizing the "oh, well, I guess it can be done!" aspect 
No biggie, I'm just glad it works as I wrote from the start... give a crystal-based Arduino good code, and it'll give you good time! Only thing to worry about is the millis() overflow every 49 days 
BTW, here's why millis() is so accurate... millis() runs in TIMER0, which overflows every 1.024 milliseconds on a 16MHz clock (every 16,384 clock cycles, with 256 values and a clock/64 prescaler):
SIGNAL(TIMER0_OVF_vect)
{
// copy these to local variables so they can be stored in registers
// (volatile variables must be read from memory on every access)
unsigned long m = timer0_millis;
unsigned char f = timer0_fract;
m += 1; // add 1 to timer0_millis
f += 3; // add 3 to timer0_fract (24 >> 3 to save bits)
if (f >= FRACT_MAX) { // if timer0_fract >= 125, add another 1ms since all the 0.024's added up
f -= FRACT_MAX; // but don't lose precision.
m += 1;
}
timer0_fract = f; // apply all we've counted so far
timer0_millis = m;
timer0_overflow_count++; //used to count cycles since this is incremented every 16,384 cycles
}
That's from wiring.c, modified with the values calculated in the macros, and with my comments describing what it does
And millis() just pauses interrupts, reads timer0_millis, resumes interrupts, and returns timer0_millis.