“But their biggest surprise came when they started adding random links to an otherwise regular network. The average path length didn’t just fall, it plummeted steeply.”
Which reinforces the idea I’m getting that we need unpopular links as much as we need popular ones.
A lot of new systems are being built that create value by automatically creating related links. They often allow for automatic reinforcement of popular links (by doing things like ordering by popularity) and as such can create problems by locking users in in a small world of connected websites, while limiting serendipity.
So you could argue that high value low popularity links are a lot better than high value high popularity links, since they have the additional function of promoting serendipity.