Gianni,

Perhaps using a 4.194304 Mhz crystal and a prescaler selected to allow you to
complete your total instructions cycle once and then poll for the timer0 is =
0.
You will need to ensure that your code for any case will always fall within a
given N * timer0 rollover.   This can be ensured by specifically placed NOPs
in many cases, and effectively can be looked at as a "agumenting the timer0"
This will allow you to have exact timing from timer0 to 0 and thus allow a
more
accurate initial approach.   The next thing I might mention is to use a 20 ppm
crystall with a tested drive resistor to limit the crystall m/uW drive.   This
allows
some release from Vcc and noise.   Using low drift capacitors (fixed) or make
one leg capacitance to ground with a variable cap or fixed and variable cap
combination.   The lower pF variable caps have a lower cap shift to temp
ppm variance. (on the xtal to gnd.)

FWIW  good luck

Eric Borcherding