Medved, I'm curious what is the maximum possible conversion efficiency for a light source which mainly functions on exciting an intermeditary element which is the case in incandescent and gas discharge sources?
I didn't understood what types of light sources you mean, when you included the incandescents.
But generally speaking, efficiency of light sources is very tricky question. Similarly the answer for "what maximum efficacy could be ever reached".
Everybody knows efficiency can not be 100% and above, that's clear.
And expressing an efficacy of a radiator with given spectrum is quite clear as well (the standard gives clear math).
I may say "all radiation is light", but then one lamp would have nearly 100% efficency and CRI100 all at the same time. Wonder which one? An ordinary vacuum incandescent lamp. Because of the vacuum, the power has no other way to leave the filament, than via the radiation. So once we count all radiation as "light", we have nearly 100% conversion efficiency. And because it is (nearly) black body radiator, the CRI is 100 as well, by the definition.
So clearly counting all the radiation is not the correct way of expressing an efficiency of a light source.
But the difficult part of your question is, which radiation count as "light", so as "useful output power", and which doesn't, so should be counted as part of the "losses".
The definition of "visible is light, not visible loss" have one problem as well: What is visible an what not? Because the eye does respond to all of the radiations, but for some the sensitivity is just very low, but not zero.
With 10..1um wavelengths this seem to be ridiculous - the eye sensitivity is so low, it does not affect any light perception, but the near IR till deep red area is the main question. It does not contribute to the lumen output, but it plays quite significant role in the color rendering. So it makes the "energy efficiency" figures of the same 14lm/W incandescent varying between 4..10%, depend if you count just the part of the spectrum representing majority of just the lumen output (e.g. 99% lumens as the "cut out" criteria), or whether it does include the effect on the color rendering (CRI99 or similar criteria).
The above appear as related to just an incandescent or continuum spectrum sources, as with usual discrete energy transition based sources the lines are so far apart, the distinction is not that difficult.
But your question was "what efficiency could be achieved". Well, we said when counting all radiation, 100% was already achieved.
But when counting just the "light" and assuming every light source will have conversion somewhat below 100% and it's CRI will be below 100, there is still quite a large room for a "spectrum shaping".
Someone calculated a 5600K blackbody spectrum cut for 90% CRI yield ~330lm/W. But no one said if you shape the spectrum a bit differently, you can not get even higher efficacy than the 330lm/W (boosting the green a bit, but extending the red/blue limits further).
This room for optimization becomes even larger, when you lower the limit for the CRI.
Today the energy efficiency of blue LED's is about 50% in mission operation (so really no sterile nor cryogenic lab environment).
As this is the only known way to directly convert an electrical energy to light, it has the potential to attack the 100% mark.
In fact the main PN junction in virtually all LED's is already nearly 100% efficient, all the losses come from getting the electricity there and radiation out. The problem is, a good electrical conductor does shield the electromagnetic radiation. So the main focus is on how to feed the electricity to the junction without obstructing the generated light. The way out appears to be in shaping the conductor so, it forms at least a waveguide for the light, there a good conductor yield lowest losses (at the end, when a good conductor does not allow the light to pass, at least it could be a good mirror).
Longer wavelength LED's are still worse (for the same absorption the resistivity should be higher than with the blue and the holes in the waveguide larger, both lead worse current transfer and/or higher light losses), so I expect more improvement there. But the hypothetical limit (when finding a perfect electrical conductor structure completely transparent for the light) is the same.
The phosphors are already approaching 100% quantum efficiency (so for each photon absorbed nearly always yield the converted one), so that can't improve significantly anymore. But as the converted photons carry lower energies, energy wise the conversion always means energy losses, depend mainly on the wavelength ratio. When converting the blue to supplement white it does mean about 60% efficiency limit.
Today the losses associated with phosphor conversion are lower than the difference between blue vs other LED chips, the concept of the blue to orange conversion will remain. Don't forget all the tricks to improve the LED efficiency will improve all colors at the same time (well, unless the blue will hit the 100% limit), so the phosphor concept will still remain for some time.
