In terms of heat:
class A => hottest but otherwise the most linear/lowest distortion
class A2 (tubes only) grid current may flow in the tubes during part of the waveform but the tube is conducting through the entire wave cycle. This allows for more power but has greater requirements of the driver circuit.
class AB => less heat, slightly more power but slightly more distortion
class AB2 => (tubes only) similar idea to A2, but the tubes stop conducting through part of the waveform. You get a lot of power and a lot less heat, but crossover distortion is more pronounced.
Class C => not applicable to audio
Class D => so far, for practical applications so far is transistor only. The devices are either fully on or fully off, avoiding the much greater power requirements of operating the transistors in the linear region. This makes the most power with the least amount of heat. Distortion can be very high, but this is a developing field, and is likely the area with the most potential for improvement in the next ten years.