It seems to me that the purpose of numbers over character codes is to constrain the available identifiers to force you to define a fairly small set of broad areas (you only have ten) and drill down within that. When we release that restriction and go with categories I think that would lead to making more and more categories with fuzzy overlap -- and that overlap is precisely what you don't want in a filing system where (unlike labels or tags) each item must have one and only one location.
You went on to mention labels and tags as a solution there -- but the point here is to implement the structure in a common filing system, isn't it? Your point works fine for email, but for documents on my drive tags don't seem like a great approach because at least on my machine even though tagging is an option, it's not robust and reliable.
It didn't feel totally fair to say that the author needed to address search more; the ability to search doesn't go away when you implement a system like this, it's still a powerful tool available to you if you know what you're searching for. If you don't know, then a system like this is a good way to explore the topics and assets in your file tree in a way that allows you to discover collections of related files.
Lastly, true that you can assign a unique ID -- but again, isn't that introducing and solving a different problem? Uniquely addressing every file might be a really useful thing, but as you mentioned, there's no context clue to help you decode B75AE2 -- you would need a lookup table. Even then, presumably you would want to know which categories and families that file should belong to.
Overall, everything you listed here is thought provoking and I think points out some of the limitations of the way our digital file systems emulate physical ones, but this collection of objections isn't persuasive that the Johnny decimal system isn't a good idea.