What PeLu said!
And a word or two more of explanation:
OK. Think of the eye. Spectrally, it is not an on or off kind of thing. At the lower red and upper violet ends of things, the eye just becomes more and more insensitive to the same amount of light power at those higher and higher or lower and lower frequencies, until not even the most powerful beam of ultra-violet or infra-red light will cause a visual sensation. So what to do? Do you simply arbitrarily define an upper and lower cut-off and say that between those frequencies a mW of light power is equal to 200 lumens? You could do that, but it would not be very good at relating what people actually SENSE. A mW of yellow-green light seems a whole hell of a lot brighter than a mW of violet light. So there is what is called the v-lambda curve which is derived directly from research into how sensitive the average eye is to varous frequencies of light. And at 555 nm, it peaks at 683 lumens/watt.
But, very few lights sources put out a single sharp frequency peak like that. LED's can, and sodium arc lights, and so on, but the color rendering of such a light source is poor. Enter the CRI, or color rendering index. 100 being perfect color rendering. Because the sun is what is called in physics a "black-body" (stupid physicists, don't they know the sun is yellow?) and because that is the standard against which color rendering is measured, ALL black bodies have a 100 CRI by definition. So that means that the most miserably yellow dim mini-mag has a 100 CRI --simply because it has a continuous black-body spectral emmision.
If you could get an incandescent filament up to 6000+ degress Kelvin so that it peaks at 555 nm (instead of in the infra-red where tungsten filaments peak), then you could get 130 or so lumens/watt, with 100 CRI. (and lots and lots of dangerous UV).
Using three color LED components to make "white" light, with a CRI of 90 or so, you could easily beat this luminous efficiency, but I can't remember the exact number. Something like the 200 mentioned. But ONLY if you are willing to settle for a lower CRI.
AND, keep in mind that many, many statistics like to confuse the psycho-physical and the physical unit systems, by saying something like "this LED is 65 percent efficient at converting power to light". What they mean is that of the electrical power flowing into the LED 35 percent turns into heat in the die, and the rest turns into light. But that DOES NOT mean that you take 65 pecent of 683 lumens/watt (!) That would only be true if the LED were emitting all light at 555 nm.
In short, it is very much more complex than just a single "best" efficiency number.