I don't think immersing the probe in a liquid other than that which is fermenting is a good idea. You want the probe to respond more quickly so hanging in the air or next to the fermenter is better.
If we’re talking about fermentation (which the OP didn’t address), then first choice would be to measure the temperature of the actual fermentation. See below:
http://morebeer.com/view_product/6666//5_Gallon_Hood_Thermowell_also_3_and_6_gallonhttp://morebeer.com/view_product/16672//Stopper_ThermowellWhen you're measuring the actual temperature of the fermentation (to Euge's point), you're increasing the control accuracy, and narrowing the variance between actual fermentation temperature and the temperature at which the fermentation environment is controlled. This point we agree on.
If you can't measure the actual fermentation temperature, I think it’s better to measure a medium that is characteristically similar to the fermentation (i.e. water). Yes, this can present some logistical challenges, but provides a better buffer so as to not short-cycle your compressor, and is a better proxy for the
environment in which you're fermenting.
Measuring the ambient air temperature of the environment would be third. As
mtnrockhopper described, it’s best to insulate the probe if you’re measuring the ambient air. Using an un-insulated probe to measure the ambient air temperature increases the risk and likelihood that the probe would sense rapid temperature spikes (think opening the refrigerator door) in the fermentation environment, and could result in short-cycling and lower than expected
actual fermentation temperatures.
While accuracy of temperature control and probe response are important, maintaining the health of your equipment is (IMO) more important. The value of my time spent in setting up and configuring equipment is (in general) worth almost as much as the equipment itself.