Yahoo! has just announced that they will be completely shutting down GeoCities later this year. I was a very early user of the service, and there are a couple lessons I've drawn as an observer over the years.
First is the danger of forcing advertising on your users. To me, this was the start of the slow decline, when the early adopters all moved on and started looking for other services.
But second is the incredible staying power of a large user base. While "all the cool kids" moved on by the late nineties, GeoCities still has an Alexa rank of 149 today, even though they've been providing a mostly sub-par experience for quite a while.
It's one of the best examples of how slowly users can evaporate once you've acquired enough of them, or in non-web speak, that people are resistant to change: once they've found a good way to do something, they tend to keep doing it that way for a while.