There is absolutely no doubt that rendering has huge potential. Anyone who has seen the likes of Toy Story, Jurassic Park, The Mask and even Batman Forever (how many people spotted the CGI Batman without being told?) will know that rendering can give truely spectacular results. But the first question we must ask, and the first of out possible "reasons"; is this sort of imagery useful in demos?

Graphics (gfx) in demos has always been divided into two easily identifiable catagories; hand drawn bitmap images, and routine effects.
Rendering is competing with the former of these two possibilities, so to have a chance at being useful, any rendered image needs to be as imppressive as any good hand drawn image.
And we see right away here that rendering has a problem. There is no dismissing the fact that rendering CAN produce equally stunning images, but the work required is far greater. Simple geometry placed on a black background and rendered with full reflections makes a nice picture, but it's never, EVER going to compete with a lovely hand crafted bitmap image.
To do the rendering genre justice means the artist needs to think way beyond "cliche" images into the field of photorealism. Imagine, for a second, that one of the movie frames from Toy Story was entered into the gfx competition at The Part 95. Toy Story is an example of "good" rendering, so it's not hard to see now why scene-style rendered images don't do well in gfx comps.

So here we have what may be considered possible reason number two; artists who use rendering for scene gfx are inexperienced or have no vision of what makes a good image.
Since rendering is very easy, there's probably quite a lot of users out there who honestly don't have the "artists touch" to make good use of the medium. While this is true for all forms of art, it's particularly evident in the rendering field, and this thought is backed up by the endless "mirror sphere on checkerboard" images which everyone seems to produce. Thats not to say they'll never learn - everything can be learnt - but it's still a problem, at least until the scene gains rendering experience.
Unfortunately, during this learning period, the output from an unexperienced render artist can be huge (as mentioned previously, it's not difficult to place a few objects in a scene and press render), and so the demo scene is subsequently flooded with primatively rendered logos and basic images. This, we can probably assume, is why rendering has such a bad name in the demo scene at this time.
But to an experienced artist who is playing with rendering for the first time, another interesting problem arises. Rendering has no "middle quality". Thats is, any rendered image tends to be either very basic, or very good, and the difference between the two seems to be base largely on how much you know, and how much you don't know.

There are so many tricks to learn which help produce a good image. But more important than any rendering tricks is the ability to see deep into what you are producing.
In movie credits, we always see the "producer" credit right at the top, but how many people know exactly what he does? Basically, it's his job to make the movie look as good as possible. A good producer has the ability to look at anything and instantly recognise it's failings, it's strong points, and most importantly, where it can be improved.
This quality is what 3D graphics artists need to aquire. The range of skills include knowing the ins and outs of lighting, colour balance, positioning and arrangement, movement and detail. Without knowledge of at least a majority of these key areas, most images will be doomed to nothing more than being yet another average 3D render.

Balance is the key word in composition, but it's also a hard thing to give a finite definition to. It's one of those things you have to "feel" as there's very few rules to use. Think about where the viewer is looking at the scene from - do you want him looking up at the scene or down at it, straight on or with a tilt - and concerntrate on how much space each item is given relative to it's importance.
A trick often used in Renaisance painting was to orientate background objects so that lines on the objects, or the objects themselves, all pointed at the most important point on the painting. In a portrait, for example, this would be the figures eyes. The effect of this was to draw the viewers eyes to where he should be looking.
Once you have a rough composition arranged, sit back from the screen for a while and really try to take in what you are seeing. Good balance can often mean even the slightest movement of objects.

Object and scene detail are the two most neglected factors in rendering. In "life", every object, no matter how insignificant, has more detail than you could ever imagine. Every surface has a bevel, and ever surface contains further surfaces.
Avoid surfaces that meet without a bevel of some degree. Even the smallest bevel can increase a renders quality by an untold factor. The more detail modelled into an object, the better the results. It's a simple calculation, yet rarely in scene renders do we witness the sort of detail required to give convincing results.

Of all aspects of rendering, perfect lighting is singularly the most difficult to achieve. Not because there are so many way to light a scene, but because of the true difficulty in recognising what makes good lighting. It's very easy to place a few lights around a scene, but on viewing the results with an untrained eye there's no easy way to tell if the optimum arrangement has been achieved.
The only way, of course, is through continued practice and learning. Lighting is an art form in itself, based largley on gaining an eye for what looks good.
Think back to a movie you saw recently, and compare it with some home video footage you can recall seeing. Neglecting image quality for a second, just how does the movie have such a great "look" compared to the video footage? The answer is, of course, lighting. The type of lens used will play a part too, but the real difference is in the lighting of the scene being shot. In the production of a movie, EVERY single shot is plainstakingly lit even when natural lighting conditions could easily have been used. For home video footage, however, rarely does lighting thought extend past whether or not there is -enough- light, and the results show this shortsight..
Rendering is the same. The "video" case is where the artist employs what is usually the default lights; a strong directional light placed over the left shoulder of the camera, with fill provided by an ambient light. This lighting arrangement is perfect for previewing a scene - it provides highlights in a place where they can be useful, and the ambient lifts shadowed areas. This, however, should NEVER be employed for a final render.
Ambient light, for a start, should be non existant, or at a very low setting (certainly no more than 5% in most cases). The reason is simply because ambient lighting, in most cases, means zero shading. The result is flat colours on objects which have no direct light, or are in shade. Shading is what gives an object a sense of that 3rd dimension which is essential given that the image is truely restricted to a 2 dimension representation.
For lighting a whole scene, you often have to fake a radiosity effect and so don't be surprised - or afraid - to use lots and lots of lights.
Radiosity is the term given to the way natural light bounces off every objects producing subtleties that rendering cannot achieve. If, for example, you have a character standing on a red floor, in reality there would be a small, yet significant, bounce of soft red light from the floor to the character. Since current rendering programs do not produce this bouncing effect, we can gain a lot by placing a similar red light under the floor looking at the character. For an example of excellent 3D lighting, go and watch Toy Story again, and then again, and again...

There really are so many facets of rendering that the task of producing a 3D masterpiece is quite daunting. Certainly nothing has been produced on the scene that can come even close to rivalling the best bitmapped images, but it's possibly just a matter of time.