>category, a subcategory, a sub-sub category, then classed 4 ways.
>attributes common to each level and, finally, attributes that will only be >needed with certain combinations of categories, subs, subs-subs and classes.

Are the sub- and sub-sub- categories exclusive or common across parent categories? The answer will greatly affect how you model the data.

Remember the Venn diagrams from math class? Conceptually, i'd think about each category as a set, and the shared attributes as a common subset. The problem is that you'll probably have so many categories and attributes that you'll need an incomprehensible hypercube to represent the whole mess.

What kind of queries will they be attempting?

One other point to ponder: Are we overthinking a simple solution?