Well, I have come up with an idea to get as much usable light off of a 200 watt bulb
as we get off of a 400 watt bulb. And then I can add a second light bulb that can be
turned on for additional brightness. It utilizes much more of the light with less of a
light density gradient across the image.

Of course with every material that the light hits, there is an efficiency loss. But with
this design, we are utilizing a band of light 360 degrees around the bulb, with the only
the light out the tops and bottoms getting wasted.

The lenses in this setup would most likely have to be some sort of glass for heat purposes
they would be cut into 8:5 rectangles and they would have a focal length equal to half the
long side. This would allow them to go edge to edge, forming a ring around the light.

There is no lens at the front, because you would have a collimator behind your lcd with the
same focal length restrictions as your smaller lenses.

We use cold mirror's in this design to dissipate heat, but if we can replace a couple of them
with hard mirror, then we should to avoid the efficiency hit.

I probably won't make this until I make my CNC machine. I would build all the mirror/lens
mounts into one unit, insuring alignment.

A possible variation would be to never have the paths reflect back through the bulbs, which
would require several more mirrors, and several more beam splitters.

Also, it may be possible to utilize the light escaping out the top and bottom, the trick would
be using off axis lenses, but you would want to make sure the ray density matched or
complimented those off the primary reflectors.

Ok, here it is. I drew in some basic light paths, I am using gimp and it is gimped. I can do the
3d stuff when I get back to my home computer but this gets the idea across, the faded lines
are secondary light paths from the beam splitter/combiners. But it all ends up going out the front.
(minus efficiency losses)

I'm trying to make sense of this diagram...

The blue objects are... lenses? With the lamps at the focal points? Basically collimating lenses then.

The grey diagonals are beam splitters and the pink lines mirrors? That's how it seems to me, anyway.

In the end though, I don't see how this will increase the amount of avaialble light beyond what a good spherical reflector will do.

Assuming that the alignment is PERFECT, and if you want to assume zero losses in the beam splitters, then maybe I could see it, but real-world devices just aren't that good. To make matters worse, you're still making the (flawed) assumption of a perfect point-source light, when in reality we have no such thing, and in fact the inexpensive lamps that make DIY so attractive in the first place tend to have larger light sources than the expensive commercial projector lamps. When you start figuring in the much higher losses for the slightly off-axis light from the ends of the light source, I feel that this will have a lower efficiency yield than the typical reflector, and at a great cost in complexity.

(I'm not trying to discourage you. I'm trying to make you think about the ramifications, and if you can come up with a solution.)