I was just sitting here thinking....shouldn't we have this by now?

So I'm curious what everyone's thoughts are on this topic.  What is preventing us from having easy-access photo-realism at this point?
What are some possible solutions to the roadblocks?
When do you think we'll break through the glass ceiling?

I'll offer up some of my thoughts.  Models have enough polys.  Texture maps are high enough res without killing processor/memory.  Those are certainly two very important things.  

How about one-touch, easy to understand realistic lighting?  No parameters to set, or very few.  No having to link them to shaders or anything ultra-techy like that.  It goes without saying that realistic lighting should produce realistic shadow-fall.  Radiosity, specularity, all that stuff should be assignable based on simple, real-life conditions.  For example:  Object is stainless steel...here's your setting.  One-touch and it's set.  

The other angle is the render engine itself.  We know from other software out there that realistic renders are not only possible, they are being done all the time, but the trick is you have to know an infinite amount of variables and check off endless boxes.  Rendering realism still takes an enormous amount of time, which leads me to believe software just simply isn't making good enough use of processor speed.

those are my thoughts, am anxious to read yours.
-WTB