Martin, what attenuation factor were you using in your probe... 10X or 1X ? I just made a measurement at 300 Khz, with a 10K resistor and just the probe itself as a capacitor... And this is the result:
http://postimg.org/image/qtbupukdr/ (500mVpp signal, 63.2% is 300mVpp, that's 3 divisions) and the calculated capacitance (10X probe) that I obtain is 14.4 pF... Which is very close to the actual input capacitance of the probe in 10X mode...
If you were using 10X, you should have obtained about 20pF... 56pF is way out... There's a problem somewhere...
The problem with unwanted stray capacitance will get worse the higher the frequency will get.
And the higher the frequency, the more your coaxial cable act like a transmission like... (beginning when 1/10 wavelength of the signal becomes comparable with the cable length) And then you want to set the output impedance of your function generator to 50 ohms, have a cable with a characteristic impedance of 50 ohms and also have a termination resistor of 50 ohms...
Keep in mind that a squarewave is composed by a sum o sinewaves at different frequencies (
http://www.youtube.com/watch?v=eC36AqL5mw8) and you must take the harmonics into consideration as well...
Of course, this method isn't an accurate method of measuring capacitance anyway... A more proper method is to to make use of the relationship between the capacitor voltage and current: (I=C dV/dt)... C=I x dt/dV ... For example, you supply the capacitor with a 1ms pulse of current, you measure the increase in voltage on the capacitor and calculate the value of the capacitance with the forumla above...