File sizes have actually been getting steadily bigger, as compression algorithms are more effective on a large document and the sum total of the data set requires less storage and is faster to query. People storing lots of data on one hosted instance, with one entry point, save money, RAM, computation, and human effort, if they keep the files large.

But a distributed, eventually-consistent storage system, like Nostr, has the opposite problem: it benefits for having lots of little files that are spread around, slowly and thinly, and then gathered and assembled, quickly, when used.