080624-1715 EST
AK:
The amount of training you have had in electrical circuit theory will determine your general ability to analyze problems like this.
This discussion points out the need to have a basic understanding how components work, electrical circuit theory, and the code and why certain requirements are in the code.
The code and certain specifications are not my speciality. So the following may need to be backed up by your study of the pertinent original source information.
I believe the specification for a GFCI is --- if a steady-state unbalanced current thru the GFCI is greater than 5 MA for 7.5 seconds the GFCI is to trip. It is also true that for short time periods this current can be much higher without tripping the GFCI. For exaample 100 MA at 0.1 second. However, an actual unit may be more sensitive. These points are on the UL test limit curve.
See my post #32 at
http://forums.mikeholt.com/showthread.php?t=99714&highlight=national+semiconductor+gfci
and page 4 of
http://cache.national.com/ds/LM/LM1851.pdf
for a trip time curve.
Fundamentally the GFCI device measures the difference current between the hot and neutral pair of wires running thru the GFCI device. It does not measure ground current, and therefore it fundamentally does not require the EGC (equipment ground conductor). Really it is not a ground fault detector but a detector of any difference in current between the hot and neutral conductors. Generally this unbalance is likely to result from a shunt resistance from the hot line to earth or some other grounded conductor, but unbalance could come from some other source than a short to ground.
120 V applied across a 24,000 ohm resistor will produce 5 MA. My normal hand to hand resistance is about 500,000 ohms. Under these normally dry conditions I will not trip one of these devices, but I will get a substantial shock. 500,000 ohms would only produce a load current of 0.2 MA with 120 V applied. However, note in a sine wave AC signal the peak voltage of an RMS 120 V signal is 120*1.414 = 170 V. Thus, the peak current would be 0.2*1.414 = 0.28 MA. But the time constant of the GFCI is long and therefore it probably does not sense the peak.
I have not tried to analyze the National circuit to see if it takes the current peaks as a measure, but most likely it is effectively an averaging device of the difference current.
If you have EGC conductors at these outlets that are really not EGC conductors because they do not terminate at the main panel ground bus, then there might not be any current flow from a hot wire to the socket ground pin.
If there are no conductive paths from your sink to the main panel grounding bus, except the water, then assume that the water in the PVC pipe does conduct to the bus, then we are concerned with the water resistance thru the pipe. I just measured the resistivity of my city water which is fairly hard at about 2000 ohm-centimeters. If the inside cross section of the pipe is 1 sq-centimeter, then for each centimeter of length I add 2000 ohms. 20 ft of pipe is 2.54*12*20 = 610 CM and thus the resistance is about 1.22 megohms. This won't trip the GFCI.
To test the GFCI you could get a 22,000 ohm 1 W resistor and connect this between the ground bus at the main breaker box and the hot output terminal of the GFCI. Using a 5% tolerance resistor you should be below 24,000 ohms and thus trip the GFCI. Using an 18,000 resistor should certainly trip the device.
I hope this gives you some idea on how to analyze your circuit. Note: soft water would have an even higher resistivity. Some bottled spring water was 2900 ohm-cm, and a different brand of bottled water was 2000 ohm-cm. A small amount of lemon juice added brought the 2000 down to 1700. Distilled water should approach infinity.
.