There is a school of thought, which says that folder arrangements are not the best way to express classifications of subject and content, when you have a database involved.
It is in the nature of such classifications, that we reconsider them from time to time, or that our needs change. So if we have used a "New York" classification, it might later be that we want to call it "NY" instead, and we may want subdivisions of Manhattan, Brooklyn and so on - or we may think we need that, but later find out that it causes more trouble than it is worth, and we are more interested in separating New York pictures into urban / interior / fashion / portrait (or whatever).
We may suddenly decide that some other major classification such as Business and Personal, is more fundamental than the use of any geographic division; which creates a conflict if folders are our method.
Folders are not flexible enough to introduce different classifications that cut across other classifications; they have to be arranged one way OR the other way. Change - which is the one thing we can confidently predict about the future - means that images will then need to move around. Some that were together will have to be separated, and the continuity of our backups (both physical file backups, and our backed-up LR Catalogs) will be compromised (or at least, disrupted) accordingly.
This is exactly the kind of kerfuffle (if you will excuse a Britishism) - tediously messy and overcomplicated situation - which Lightroom was intended to avoid. Changes of Collections membership and keywording do not involve any moves or changes in the storage of the files, and different kinds of classification can freely intersect with or completely ignore each other.
Each image can then enter the computer filing according to some STABLE FACT about the picture which is never going to change - such as the capture date - and be stored according to that in a subject-neutral system. One such system, is YYYY/MM/DD. Once stored, the images will remain in the same place longterm - or at least, withih the same structure; even if that moves partly or in full to another drive in the future. We then have no need to care what the folders are called - we just want it systematic and reliable. Every image has just one unambiguous date that it was taken, with no room for interpretation, and can go into a fully predictable location based on that.
There is no reason why a richer range of folder systems would not be of benefit in LR; perhaps a more user-programmable one such as is found with the Filenaming dialog. But the more one looks to achieve subject or content classifications within that which are not inherent in the images, the more complex and the less feasible automation becomes.
