@randombit0
I also did an experiment with a quartz tube and a Kanthal heater coil. You may be interested in the results, although not very well constrained overall.
The quartz tube was heated to a set temperature (516°C), and radiant power calculated by subtracting total input power W from the calculated convective loss (by standard formula). ε was determined by comparing to a thermocouple and correcting an IR probe ε until T matched.
Power by ε was calculated by the standard formula, and compared to the power calculated by subtracting convection from power in.
Then the quartz tube was painted barbeque black, and retested at the same input power.
Temperature increased almost 10%, but this merely increased the convective losses, so that total radiated power remained the same.
Within the admittedly large possible experimental errors, defined (by me) as the difference between a ε of 1.0 for total power output plus convective losses and the necessary ε to get the power to balance (0.96 vs 1.0), the radiated losses unaccounted for by transparency, due to the spectral sensitivity of the IR probe (8 to 14 microns), was about 10% . This is the 10% that was converted to convective losses but a higher temperature.
Therefore, although flimsy evidence (but could be repeated for better confidence), in open air, a selective emitter is a more effective radiator of electrical power converted to IR than a blackbody, since the blackbody ends up with a higher temperature and therefore higher convective losses.
This is, of course, below the temperature where radiant emission exceeds convection by a much larger percentage. At, say, 1000°C, this might not hold up as well.
In other words, perhaps 10% of the heat energy could be "lost" (remain unmeasured/calculated) in the dummy. But as the alumina begins to become incandescent, the transparency decreases. In the Manara et al plot, the maximum temperature is about 777°C, which is just at the tipping point of radiance overtaking convection as the primary heat mover, based on my experiments. Another trace at 1273K would have been very helpful.