If it’s discovered, I’m real surprised it hasn’t been noted somewhere,
because original values/calculations are still being used also.
It would be affecting commercial products out there I’ll bet.
If you assume 7 instruction cycles for reload, and use the correct overflow value, isn’t that still exactly the same problem?

Test is easy.. make 10,000 Hz hardware PWM, and 100Hz PWM with Elapsed timer on the same chip.
10,000Hz is the lowest even freq at 50% duty cycle without resolution error assuming 10 MHz clock.
So then it only takes one pic and two dividers running from the same crystal.
Then any old crystal of any value doesn’t matter... run the chip as fast as possible, until the LEDs run away from each other.