Ryan Singer wrote an interesting post titled "Features are a one-way street". The main idea was that "once your user base has grown beyond a certain point, you cannot take features away from them. They will freak out. Whether the feature is good or bad, once you launch it you've married it."
Google is a company that launches many features and services "in beta", tests them for a while and then decides whether it's a good idea to continue developing them. Some of the Google Labs projects were successful, while others were discontinued or they were integrated in other experiments.
In a talk about innovation, Marissa Mayer mentioned an idea that came from Eric Schmidt: "don't kill projects, morph them". I don't know how many people miss Google Voice Search, a service that allowed you to search Google with a phone call, or Click-to-Call, a simple way to call businesses for free directly from Google Maps, but both services are the predecessors of GOOG-411. The cool Google Deskbar was partly integrated in Google Desktop, the SOAP Search API was replaced by a REST API, Google AdSense Referrals was superseded by the DoubleClick Performics Affiliate Network, while a replacement for Google Answers is being tested in Russia and China.
Other products were discontinued without offering a replacement: Google Video Store, the photo sharing service Hello, Google Browser Sync or the "Send to SMS" feature from Google Toolbar and from Google Send to Phone.
For Google, features aren't a "one-way street" and you never know when a functionality you start to rely on is removed for unknown reasons. That's why I think Google should label experimental features more prominently and it should do a better job at communicating the reasons why a service is dropped.