warpie said:

What is the criteria for choosing the correct coupling cap in a circuit.

I realise that the value depends on the impedance of the circuit and on the required Fc, which should be well below 20Hz (is around Fc=5Hz low enough)? However, I've seen circuits where (let's say) a 10uF cap will provide a cutoff @ 5Hz but instead a much larger cap is being used (22uF, 33uF or even higher).

In jurassic times, engineers aplied the rule of 10, i.e. they calculated the value for the desired -3dB frequency and multiplied by 10, because when you cascade 10 stages, the resulting cut-off is about right.

In addition, since electrolytic capacitors tended to degrade considerably, they took a margin.

Today, the quality of electrolytic caps has improved considerably, but we are also more concerned with other aspects than basic frequency response, particularly distortion.

In order to minimize distortion, the voltage across the caps must be kept very low. I apply the rule of 100, so I calculate the value for 0.02Hz, which results in 80uF for 10kohm load, that I round to 100uF.

For a 600 ohm load, it results in 1500uF. Hower, due to the bulk of such a cap, I prefer using a THAT1646, that uses small bipolar 'lytics.

This was almost impossible in the 70's but today, the size of electrolytics has decreased enormously.

Film caps' distortion is at least an order of magnitude lower, so you can safely apply the rule of 10.

I somehow feel that ESR and self resonance play a role here

Not so much. ESR just adds a small loss and self resonance appears well beyond audio frequencies.