I used RobotLegs for a game that required integration with a few different backend services. One service gave me a multi-player lobby area to allow players to challenge each other, one handled the head-to-head gameplay after a challenge, and one allowed players to access information from their social network accounts.
From the beginning, I planned the game's architecture as if I were building a Rich Internet Application. The game itself was an implementation of a popular turn-based board game. Thinking about how to run a local game vs a network game definitely helped me stay on track with the MVC framework approach to game development. There was a ton of code that could be reused, and the difference between interpreting a local player clicking with their mouse versus receiving a message through the network to indicate that a remote player did something similar helped me see what logic simply couldn't be tied to the view at all. I was able to use models, commands, and mediators very smoothly, and in the end, it made the game's code more maintainable and easier to understand when I delivered it to my client.
I think most games will have a basic model that tracks the "board" whether it be pieces in a grid or enemy ships and asteroids out in space. Once you're thinking of the model as a separate entity from the view, it should be easier to imagine how player interactions through mouse and keyboard could trigger commands from controller to make changes to the model and notify the view of those changes. For some simple games, that might end up being a lot more work. For others, such as those where long-term maintenance or multiple input methods are required, it can save some headaches.
Let's think about the different views in a game for a second. Views can include a title screen, a settings/options screen, a multi-player lobby, high scores/leaderboard screen, and the main game itself (which could consist of many smaller views!). Many of these views can have models, like the list of high scores, the various settings (which should be shared between the options screen and the game views), the list of players waiting for a game, and the current state of the game, etc. By the way, need a way to save the game so that a player can restart where they left off? It's way easier to do that when the data is in a model and not tied directly to the view.
I think that too many Flash developers look at games as hugely different beasts from Rich Internet Applications. An MVC framework can be appropriate for a game, especially for multi-player and for games that you intend to iterate on over a longer period of time to add new content and features. The biggest challenge is making yourself remember the fact that cute, little furry critters running across your screen is just a visualization of data that might easily be displayed in a different way using DataGrid or a chart... though it may not be as fun to play with them that way!