I've come up against various issues dealing with this problem myself;
1) The resolution you can *actually get* out of a pot. For a 30mm knob, you've got 94mm of circumference. If you've got a 300 degree pot, you can only use 30/36th of that = 79mm. That means for 8 bit resolution, you're talking about adjustments of less than a 1/3rd of a mm. You can't really adjust that finely.
2) ADC noise, and the limits that makes on resolution. You might be able to read 10 or 12 bits, but you might well get values that move by +/-20 in either direction. I've had to use running averages to help reduce noise.
3) Table resolution doesn't have to be the same as the ADC resolution. You can always use just the top bits of the ADC result to do a lookup into your table. If you want more accuracy without using up more memory, you can use the low bits of the ADC result to interpolate between that lookup value and the next one. This is a (typical) trade-off of computing time against memory requirement.
4) High resolution in tables doesn't guarantee no zipper noise. If there are fast changes in the table, there'll still be big jumps, and that means you'll hear them. I've had this problem with tables that convert from a linear value (a CV say) to a exponential value (Frequency for example). Making things smooth is partly about keeping ranges sensible to ensure there are no massive jumps between table values.
5) The speed of the ADC readings is a consideration. 14-bit MIDI pitch bend values can still sound steppy as hell if they only come in every 200mS. The important thing is not the *resolution* per se, but the *jump* from one value to the next. Obviously if the ADC readings are much quicker, the jump gets smaller for a given rate of change.
Hope this helps with your deliberations!
Tom