I don't want to start an argument but the ADC only needs 1.6 nA to read an analog voltage. I don't know how you came to the conclusion in the above statement. Can you explain what you base that statement on ? I've used Mohm values for voltage dividers with analog inputs with no problems.
The above is mostly repeating what I've read here. I recall reading this sort of thing from people I consider knowledgeable on the subject (I don't recall their names).
I haven't tried using the Arduino's ADC with high impedance sources but I was under the impression many of the commonly used ADC devices require low impedance sources to work properly.
I know the MCP3208 has this limitation.
I believe there are things one can do to mitigate this limitation. If the ADC is sampled infrequently I believe the accuracy is improved when measuring high impedance sources. A small capacitor between the ADC input and ground can also improve the ADC's performance.
Hopefully others will chime in to eliminate any uncertainty on this topic. I'll have a look at the AVR datasheet to see if it has anything to say on the topic.
Edit: As I've thought about this a bit, I realize I should have added a caveat about sampling frequency. I'm still under the impression, the Arduino's ADC has a hard time reading high impedance sources at high frequencies.