perhaps you could implement some of the cool material features now
Then he basically has to do work which he's going to throw away later. It's not "all or nothing" so much as working efficiently vs. inefficiently.
Also the new shader system is the headline feature of 5.x so dropping it would not help with buzz.
In the current system the procedural marble doesn't do transparency or blurred reflection. Where's he going to plug in SSS? Into the standard "material"? If so you can't do marble with SSS. If he plugs it into marble, you can't do SSS plastic or leaves or flesh. The current material system comprises a flexible "material" with no procedural components, and a bunch of procedural components which aren't flexible. This is why SSS doesn't make sense in the current system. So Martin either builds a huge number of special case shaders which need to be tossed out, or he builds a new material system which gets it right the first time.
Note: the new material system is mainly of use to the archviz / industrial design types (even if they don't realize it). I'm not arguing my own case, which would be for improved animation tools.
how would volumetric fog work? Would this be possible to work kind of like a volumetric lighting, for example if you would make a light saber or a infrared light from a gun or something?
Volumetric lighting is done in a variety of ways, e.g. voxels or rendering z-buffers from the point of view of the light. The way Blender does it is by rendering a series of 2d slices of the (spot)light volume (you can see exactly how it works if you look closely). It's a kludge, but looks fine for animation (not so good for high quality stills, where having a really high rez z-buffer would probably work better).
Volumetric Materials are done by keeping track of how long a ray's path through a material is and coloring the "virtual photon" passing through accordingly. There are further wrinkles (such as projecting 3d noise into the volume to get "wisp" effects). This lets you do localized fog effects, clouds, etc. Self-shadowing volumetrics are another story... Haven't thought about how they're done but it's probably by using some kind of voxel or particle approach.
Martin might want to avoid doing volumetric materials until he can do self-shadowing because he prefers to do things once the right way rather than provide half-assed hacks ;-) (This is why c3d's renderer has fewer features but produces output comparable to much more expensive programs.) I don't have the link handy, but if you see what Martin has said in the past about how he wants to implement hair, it's very much along these lines -- do it right or not at all.
What is anisotropic materials?
Basically surfaces which have different reflective or transparency qualities based on the direction you look at them from (not just angle of incidence, which is fresnel effects, but direction, as with brushed metal).
Hmm, can you do it(easily) in Blender?
It's about as easy as anything else in Blender ;-) -- it took me about an hour to render the image I posted (including working out how to do everything). Blender's material system is really quite impressive (as far as I can tell it's actually not a full shader tree, you can just plug procedural or texture maps into any channel and it supports any number of UV and vertex color maps; but I could easily be wrong since Blender is kind of hard to grok).