The martian thing looks interesting, but not really what I had in mind.

The sort of experiment I was thinking of would involve a set of boxes, each big enough for a small plant, and designed to consume the same electrical watts but produce different light spectra. Maybe have some with 1 red 1 blue, some with 2 red 1 blue, some with added far red and far blue, some with broad spectrum white and so on. Control for temperature, atmosphere, nutrients and growth medium, and then calculate yield per watt hour for each. Of course the comparative efficiency of different LED wavelengths will probably continue to develop over time, so maybe calculate yield per photon flux aswell.

Maybe test two spectra at a time with several under each spectrum. Wouldn't take many seasons to get results with a reasonable likelihood of statistical significance. Could probably do it in a few years if I took a bit of overtime to increase the tinkering budget.

I find the whole subject fascinating, but I haven't had much luck finding data collected under controlled conditions, so it's a good excuse to solder things.