I'm still shocked at the quality and accuracy of the resonator on the board. I'm going to pick up another board this weekend and measure the resonator to see if it was just a fluke.
As lefty indicated, tolerance ratings are no doubt statistical in nature, and individual parts
will lie along a Gaussian distribution, so many will be close to the center, but don't bet
on "always" getting good timing accuracy.
If you get a bad part, and figure out the error at 0.5%, it's really quite significant, if you
want good timing in your sketches. 0.5% is 432 seconds per day, and that comes out to
300 msec per "minute". I don't find that encouraging.