Hate to bring some real engineering practice into it...
Going to have to do the same to you @fungus.
Variation is about 5% between 3V and 5V Vcc (eg. when powered with 3xAA batteries).
Variation is about 5% for every 20 degrees difference in temperature (a difference easily possible from summer to winter).
For the entire envelope I get a total
a variance of 5.3125%. Centered at 8 MHz, the variance is +1.5625% / -3.75%.
5% variation is enough to make it fail.
That's 1/2 bit error which a modern hardware receiver easily deals with by sampling an odd number of quanta greater than one and tweaking the clock.
A modern UART should have little to no trouble receiving data over the entire temperature / voltage range.