Another round of testing with an uneven Chinese heatsink and using low and high clamping power. Loose connection was with the screws just holding the MCPCB put, but not really clamping it down. Tight was with the little screws as tight as was practical.
This time the results were more interesting with the No TIM being horrible. The normal thermal paste performed almost identically with low and high clamping power while the graphite pads shined when the MCPCB was clamped on tight. In that case they were actually outperforming the paste. This could be due to their ability to conduct heat very efficiently in the lateral direction (i.e. the small hotspot generated by the emitter was spread effectively to a larger area).
Ambient temperature was 27°C. Temperature number on the graphs are delta over ambient. I measured both the MCPCB and emitter separately.
No TIM with loose clamping. LED is red line, MCPCB is blue.
No TIM with tight clamping
Panasonic PGS graphite pad with loose clamping
Panasonic PGS graphite pad with tight clamping
IC Graphite pad, loose clamping
IC Graphite pad, tight clamping
MX4 thermal paste, loose clamping
MX4 thermal paste, tight clamping
Output comparison with a 1A output for reference. First with loose clamping. MX4 is marginally better than the pads.
Identical results for the TIMs with the screws tightened
Similar story with the forward voltage. Higher heat equals lower voltage.
Loose clamping
Tight clamping
Conclusion: With less than optimal mating surfaces, if there’s not a lot of clamping power available, use thermal paste (although in practice graphite pads are similar in performance). If secure mounting with lots of clamping power is available, graphite pads are at least as good or even a bit better than paste. No TIM at all is significantly worse in both cases with an uneven heatsink (what a surprise).