However, "on average, each recording in the collection is only accessed by one researcher per month" seems to me like a dishonest way to present the data. Surely surely there are one or a few recordings in the project that are vastly more popular than others. That's always how it goes. So "on average" completely obscures how popular the most popular ones might be.
Their approach stands on applying the law as written: "This is mine and you can't have it, even if I don't really much care about it."
You might have a few hotspots, but there's a very tight upper bound on how many downloads of each item. Further, it argues that the overall collection has a very high research and preservation value compared to the total number of downloads.
Quite clearly what the plaintiffs are afraid of is that there are a few very popular works in there that they could be making real money from. There's no way to know this from the mean alone.
The companies have very successful legal departments. They go after the next lowest hanging fruit. They are just working, it's their job description to do this. This is why they will never stop, it never gets too absurd or overly unreasonable.
edit: Deleted fake example