I did have a look at the datasheets but I am reluctant to use this method as there is no way of knowing what the frequency is in the circuit, unless I carry around a very accurate multimeter.
I would be comfortable trusting the manufacturer's specifications, after doing an initial test to make sure I was programming it correctly.
Of course any device needs an accurate frequency source; if the device doesn't include one internally, you'll need to provide one, and its accuracy must meet the specification.
You mentioned the tolerance of ceramic resonators but don't crystal resonators have a smaller tolerance?
If you mean quartz crystals, yes. Have a look at this data sheet:
http://www.abracon.com/Resonators/ABLS3.pdf
I THINK (not certain) that it's pretty typical of crystals in this frequency range.
As you can see, the crystal is available in a wide range of frequencies, and in fundamental and third overtone (harmonic) versions. This series of crystals is available with an initial frequency tolerance at 25 degrees Celsius between ±5 ppm (which corresponds to about ±180 Hz) and ±30 ppm (±900 Hz). This is typical for crystals in this frequency range. As well as the initial frequency tolerance there is a temperature stability tolerance across the operating temperature range (typically 0~70 degrees Celsius) of between ±15 ppm and ±150 ppm.
You can get individual crystals in certain frequencies from Digikey, but if you want specific frequencies that aren't standard, you'll have to buy a reel of 1000 units! There is only one frequency in the range you gave that's available from Digikey:
http://www.digikey.com/product-detail/en/ECS-352.5-18-1/X151-ND/226620
This is the main reason why I suggest a frequency synthesiser approach. Digikey will sell you single units with initial accuracy of ±10 ppm and temperature stability of ±10 ppm for around USD 1 each if you choose a standard frequency like 12, 18, 24 or 32 MHz. The frequency synthesiser chip can then do the work of producing the exact frequency you want.