The crap results were probably due to the soldering messing around with the crystal properties. Using a pressure-contacts approach instead seemed to do the trick:

http://tinyrock.com/resources/4000918/Bandgap-Measurements-I...

Has it really been 10 years!