Lensed LED lights waste a great deal of light from the LEDs, but they focus the LED element very effectively at a distance, capturing the portion of the light that reaches the lens and producing a die shaped image downrange. But when you put them in the integrating sphere and measure their total lumen output you find that a great deal of the LEDs total output has been lost compared to a reflectored version of the same LED.
The reflector focusing effect with a tungsten filament is much greater (than on an LED), as only a very small fraction of the light is directly radiated forward from a white hot filament. But even with power LEDs the output pattern is spread over most of the forward half of the sphere. The flux is higher near to on-axis, but the integral of light at wide angles that hits the reflector is still a large fraction of the total. Take a reflectored LED flashlight using a power LED, observe the pattern, and then remove the reflector. Most of the central beam will disappear, the lens on the power LED doesn't make a strong beam like pattern. What's left is a really broad swath of light missing the bright long-range central beam.
The big power LEDs put out quite a bit of light at wide angles, unlike the small encapsulated LEDs that have built in lenses. Here's a graph of a Cree XP-e:
http://reefbuilders.com/files/2013/05/XP-E2-Color-viewing-angle.png
It is fine to throw away some of the light in order to control the pattern, but this loss in efficiency will either require more power or result in less range.
The real trick here is the quantity of light needed at different angles is different. Close in it doesn't take much light, at a distance it takes a lot more light, and above the horizon you want no light. It is not an easy pattern to generate. A lot of lights put out so much light on the ground in close that your eyes de-sensitize and you can't see the main beam very far out. It needs to be dark enough in close to allow you to see farther out, but not so dark that you miss important things in close.
After a lot of on-road commuting experimentation I have removed all the extra lights on my eBike and now use only the Fenix BT20. It works best by aiming it slightly down so the central beam gives good light at a distance (beam pattern just below the horizon), and the fresnel section in the upper lens cuts the upward spill that would be in the oncoming driver's eyes and directs it instead downward to fill in the darkness between the front tire and the lower spill directly from the LED die. What I find is that of the several other lights I tried only the Fenix lights up near and far without a lot of lumens, and without throwing a bunch of light upward where it doesn't belong or lighting up the ground too brightly in close. It doesn't seem super bright, but it allows good light near and far and allows seeing effectively farther than other apparently brighter lights. I ran both after adding the Fenix, then I tried turning all the other lights off and only running the Fenix. I found I could see the road debris in the bike lane farther out, and that the light was well controlled below the horizon. I kept the other lights for backup temporarily, then removed them. I carry an extra set of 18650's for the Fenix, later I will run them from an 8V regulator and do away with separate batteries. An ideal setup might be two of the Fenixes, I haven't tried that. One seems to be enough, a second would be primarily for redundancy.