Amen. After 2 years with the render tree in production, I'm a far better look artist than 6 years in other apps using other paradigms. Start with learning to understand the mixer nodes. Once you get that down, figure out texture space, how to control it, how to warp it, etc.
Most importantly, working with the render tree forces you to think in terms of compositing. In order to build your shader, you must break it down into its specific parts. Look at the overall diffuse shading. Figure out the source of the speculars... how many are there, can you reproduce them with default illumination speculars or should you do a reflection map? What are the imperfections? Scratches, dents, dirt? Then you re-assemble all the source parts into one whole shader.
Since you begin to train your mind to break things down into separate image sources, you have a natural segue into breaking out your passes for post effects. You'll discover pretty quickly which effects you can get Mental Ray to do, and which ones are going to need some help in post. And hey, look at that. The compositor in XSI can handle most of those effects without even leaving the app!
All I can say guys is stick with the render tree because:
1) I can now texture faster than was efver possible in Soft3D, even for simple things. And this while I've taught myself XSI, whereas I was trained at Mesmer in Soft 3D. Once you get the hang of it, it's a godsend.
2) You will learn so much more about shading and rendering than you would otherwise. And that will make your images better no matter what software you use in the future.
It's a bit more work to understand, but in the end it's well worth it. If you haven't tried this already, download some of the render tree presets available at edharris.com or softimages site. Apply them, then start dissecting them, playing with sliders, pulling nodes off, etc. You will eventually find amethod in the madness.
Scott