Interesting numbers! I havent as yet got the gear to measure the actual performance of my TIRs, and yes, we do see light going out the back of the reflector. I think thats due to random reflections from the parts of the TIR that arn't exactly meant to be part of the reflective optical system. (mounting tabs, rounded moulding edges, moulding sprues, etc..
The theory behind TIR relies on the interface between the optical plastic and the air. For a given difference in refractive index, if light hits the interface at a low enough angle, it will be totally reflected back into the material (thus the name Total Internal Reflection). This is how fiber optic cables work (in simple terms). In a TIR, the angle of incidence between the light from the emitter and the sides of the TIR are supposed to be within this angle that allows internal reflection, and therefore losses are only from the losses through the material (giving 98% efficacy or something) however there are factors that can cause light to be lost, again related to quality and design.
1) Optic is too small compared to the emitting source (small optics, big leds) - results in a decrease in the ability for the optic to focus the light in the case of a reflector, but in a TIR, can actually cause light to be lost out the back because the light begins to go past the angle of incidence required for total internal reflection (less is reflected, more is refracted (out the sides)). I havent explained this too well, but its because the LED is wider, relative to the reflector, so the outer edges of the LED emit light which to the reflector are outside of its ideal focal point. Ray diagrams here would help.
Just thought of an example, looking into a fish tank, you can see through the far side of a fish tank if you look straight into the tank at right angles to the far side, however if you start looking through it at an angle to the far side, you will start to notice that the far side will start to look shiny and you cant really see out of the tank. Thats light that your seeing from somewhere else that has been reflected of the inside surfaces of the tank.
2) Optical quality of interface - TIR relies in the interface between the optical material and the outside air. Any contaminants, oil, fingerprints, scratches!!, dust will alter the ability for light to be internally reflected, and will result in losses
3) Moulding quality/Design - Many optics simply have rings at the outer edge to facilitate mounting the optic, however these become involved with light from the emitter, but the non ideal shape means light gets reflected around randomly. This may account for another 2-5% of light, which you will see as rings outside of the normal light output (out the front, sides, back, anywhere)
4) Design - TIRs Should be coupled with the intended emitter (as reflectors are) because they Should have been designed to accomodate the emissive characteristics of particular emitters. Many chinese lights save money by using generic TIRs, which for us means TIRs for luxeons (or their cheaper alternatives) for newer emitters such as crees. These don't work as well because they are difficult to focus accurately, without more complicated mounting systems, and are likely to induce more losses again.
Overall there are many places where light can be lost, but the story is not to different for any optical system. Cut the budget enough and there will be losses in performance. Reflectors are also high efficiency in theory, but the reflectivity of the surface, dust contamination, shape of the reflector, all these apply all the same.