The question re how efficient actual LEDs are is a good one, but the answer is more complex than may be expected. Illumination capability is usually expressed in "lumens".
LED efficiency is usually expressed in terms of either
light energy output or
per unit of energy input.
For a given lumen output, efficiency is usually expressed in lumens per Watt (l/W) or in light energy output per Watt W/W). The first figure is more useful in practical illumination applicatios, but the second is more meaningful in terms or energy conversion efficiency.
If lumens and light energy had a fixed relationship then efficiency determination would be simple. However, What a given lumen figure represents in terms of "light energy" varies with the spectral composition of the light.
Lumens are expressed in terms of the theoretical response curve of the human eye. The same amount of light energy will produce a different number of lumens as light wavelength or mix of wavelengths varies. As a consequence, the wavelength or wavelengths of the source plays an important part in the lumens produced per energy input.
At the short wavelength end of the visible spectrum (not quite UV) eye sensitivity is extremely low, so lumens/Watt are low - so much so that it is usual to quote the output of deep blue and "Royal Blue" sources in terms of mW/W (light energy per electrical energy). This is highly useful as an LED family which includes phosphor-less and phosphor based LEDs allows some comparisons to be made.
For example, the "top flux bin" of the Cree Royal Blue XT-E LED when operated at Vf= 2.85V and If = 350 mA produces 613 mW typical (600, 613, 625 mW min/typ/max) at a wavelength of 465 Nm.
That equals an electrical to light conversion efficiency of 60.2% / 61.5% / 62.7% min/typ/max.
See page 19 of Cree XT-E datasheet top right of table - XTEARY-00-0000-
The top white phosphor version of the same LED produce 180 lumen at 25C at
2.77V, 350 mA = 970 mW DC in or 186 lumens/Watt.
IF the light energy from the Royal Blue & White LEDS was the same then the white LED would have a 100% l/W figure of 186/61.5% = 302 l/W at 100% efficiency. However, light outputs are not identical (quite) as in the white LED a portion of the LED die's blue light is used directly and the remainder excites the phosphor(s) with some loss in light/light conversion efficiency.
As has been noted, Wikipedia (correctly) states that the maximum theoretical l/W figure is 683 l/W.
How can this be reconciled to the claim that 100% white LED efficincy is ~= 300 l/W - and the fact that various manufacturer's are now making white LEDs with efficiencies > 300 l/W?
The answer lies in the useful but arcane (or arcane but useful) fact that the lumen rating is related to eye response. Maximum eye sensitivity occurs at a wavelength of 555 nm. The maximum possible efficiency in l/W is achievable with a monochromatic source at 555 nm. ANY other source, monochromatic or multi wavelength, will have a lower 100% theoretical possible l/W figure.
The "ideal" white light source is a black body radiator at 5800k with its spectrum truncated to the 400-700 nm range and has a max efficiency of 251 l/W !!!!
By making various adjustments to maintain "white" light while altering the % of various wavelengths increasing white efficiencies can be achieved.
A 2800k black body truncated asymmetrically to achieve a CRI of 95 has a 370 l/W max theoretical efficiency.
But wait - there's more, but, later maybe.
I'll come back and add sources and more detail, but the above shows that the answer is harder than the question, and demonstrates that in true energy out per energy in terms the top modern LEDs achieve energy conversion efficiencies of > 50%.
More anon - light fades - rootop job beckons ...
Analysis on the Luminous Efficiency of Phosphor-Conversion White Light-Emitting Diode