First about the setup:
I made a SCPI script that controlled the
PSU: setting the voltage and on/off,
AWG: ref voltage (Temp) and trigger signal (0.2 Hz)
Scope: measuring amplified current, acting on trigger.
The amplified shunt voltage was measured using the average of 14.0kpts every 5 seconds. After each measurement the VOffset of the channel was "centred", so the readings could be done using 50 mV/div.
The current regulation had a bit of an oscillation. Also the temperature regulation was not oscillation free. This is not really an issue because both are averaged. One using the 14.0 kpts. The other using the last 100 seconds of those averaged values.
I made a modification of R11 (100K -> 33K) that made the oscillating worse at higher powers.
The measurement where done in the shade/dark, that actually makes a difference compared to normal daylight. Also I left the room, so no incidental air movements.
The best top was as expected the large mesh. The solid top and covered side holes only performs 8.7% less.
At 80 deg celcius the top became a bit softer. I wouldn't design for that temperature, 60 deg would be ok though. For my own usage I would like it to run up to 40 deg environment temperature, thus only 20 deg difference, that would mean 1.6 W would be the average power it could dissipate. With a lot of headroom, if one doesn't deform the plastic when it's hot.
The graph doesn't show that there's a lot of heat buffer in the case. I ran the tests long enough for that. The heatsink has a little bit of buffer, and also the top, but both of them don't weigh much. The rest of the box doesn't heat up much.
But now a 8.7% difference is shown between a good top and a solid one. Not much, but if there wasn't a difference I would have found it necessary to dig further into that mystery.
What did the experiment do for my intuition?
It enforced my believing that one cannot dissipate much heat in small plastic boxes. It changed my thinking that airflow can make it a lot better.
Printing a thinner top, with surface enhancing structures (inside and outside) together with a plastic that conducts heat better might give similar or even better results.
I also learned not to just trust the dissipation values in the data sheet.