In the Memory Ward: The Warburg is Britain’s most eccentric and original library
newyorker.com
newyorker.com
So for instance when I browsing the section on early chemistry and alchemy, I found a German alchemical text from the 1530s just sitting there next to the 20th century stuff. It's remarkable.
Walking down the stairs I noticed a rather interesting looking flag in a case on the wall - it was from the Battle of Flodden in 1513. It was just mixed in with loads of other interesting stuff...
I might forget some bit of esoteric syntax and I can look it up and move on with my life, but if I'm flipping through a book (or books) to find the information, I might better understand why that syntax came about. (I'm struggling with a particular instance of this which is why it comes to mind right now.)
Accepting answers instead of discovering them... maybe it doesn't scale anymore. (Insert get off lawn joke as necessary.)
Back in college I used to use the card catalogue (or the computer, if I had to — but the card catalogue was better for serendipity!) to look up a few books on my topic, then I'd walk to the shelves and thumb through every book on the shelf above and below the book I'd found. I'd then check out every book I thought would be relevant to my work.
It turned out that this was an excellent way to research topics. Rather than assume I alone knew the words and concepts to search for, I put myself in the hands of those who had classified a bunch of books similarly — and magic happened.
I can't say I see it much elsewhere though. On HN for example, sometimes a linked article contains links to other, sometimes unrelated, articles that attract my attention, but I rarely wander off along a chain of distinct sites.