There's a problem with classification systems (e.g. DDC, LCC, etc.). And it's not a secret. I certainly didn't discover this problem but it's not going away and classification will still be important enough in the near future that it would be nice if it were solved.
The problem lies in the two pressures on the system: change and resistance to change. The resistance to change is obvious: in a library of any decent size, change means work. More change means more work. Libraries do not tend to re-classify every item that would otherwise get a new call number every time OCLC and/or LC decides that the system needs expanding or restructuring. And consistency is good for the user as well; if they've become accustomed to the system as it is, then altering the system will be a little stressful. Not too much but a little.
But on the other hand, classification systems, at least the currently used and practical ones (i.e. the ones we like) are founded on something that does change, and quite rapidly nowadays: subject terminology. The words we use to describe what books (mostly) are about, the hierarchies that we put these terms into, the disciplines studying or working in these subjects, and the relative popularity of all of these, all change over time and therefore push the classification system in the same direction.
I've always been a fan of automating where automation is helpful. (Duh.) My thought is that if classification was linked algorithmicly to something else, that would at least reduce the amount of work necessary to push through any necessary changes. It would also make it simple to apply classification to items not usually seen as being worth it as long as it had this something else. I guess my bottom line is that change is inevitable. As, I think, Charles Cutter said, these systems should be built with the ease of the patron in mind, not the ease of the librarian. If the way items are accessed should be changed based on the ease of the patron, then, as long as it is practical, then it should be done.
This something else, in my mind, is subject indexing. To an important degree, classification is already based algorithmicly on the subject understanding of the item. But it's not been done to such a degree as to make it simple for a machine to take a list of subject headings applied to a resource and spit out a classification number. But that's what I think would help. Changes in subject indexing requires a lot of work as well so this in itself doesn't solve anything but if subject indexing and updating it could be made more automated, easier and more frequently updated, then linking classification to it would apply the benefits to both at the same time.
HOW this would be done is not clear. It would probably require some alterations to systems of subject indexing as well, such as identifying relative importance between various subject terms as well as relationships between them. And there are practical considerations to make it work in the real world. For example, systems would have to be in place to alert library staff to when resources had been reclassified and how, so that physical changes to the item labeling would be made quickly and easily. Perhaps transitional changes from DDC and LCC could be managed so that, when adopting the system, libraries are not faced with a mountain of work immediately. The transition could be managed at a reasonable level.
The resulting system would make the classification system chosen less of a system unto itself but rather a dependent subsystem of something else, reducing overall workload for everyone involved.