Sorting's failure to scale is a specific example of a general problem with interfaces that don't scale well. One of my least-favorite examples is the way Apple replaced it's excellent Sherlock search interface with Spotlight.
Spotlight probably works fine for users with 1 to 1,000 data files. I'm a writer, I've owned computers since 1987, and I'm a backup fanatic. At one point my Documents folder had over 300,000 files in it. Instead of getting 10 search results like Joe User, I get 100, or 1,000. So I need an interface that allows me to
Easily choose WHERE I will search (which folder or folders);
Quickly SORT the files on the basis of what I think will help me find what I want,
Quickly SCAN the files, and
Quickly RE-SORT the files by a different parameter if steps 1 and 2 don't work right away.
Spotlight may have a way to quickly delimit which folders to search, but I haven't found it. Worse, it sorts results into seperate categories, and displays only SOME of the hits. This makes it impossible to just scan the whole list of files for a match. If what I want isn't in the first few hits, I must also remember what KIND of file it is. Ack! More mental overhead!
Mental overhead does not scale! I can't just add a faster CPU or more RAM to my brain.
I suspect that Spotlight's developers came up with their cool idea, implemented it, and tested it ... on machine with a few dozen to a few thousand files. And it worked -- in that environment. I ended up turning it off and use FileBuddy for searching. Not ideal, not as good as Sherlock, but it works.
The traffic here in the Bay Area is another example of technology that doesn't scale well. Trains scale well; autos do not.