The controller is the glue that binds the model and view together. You want both your model classes and views to have as few dependencies as possible on the other layers of your app. Therefore, the controller should always load data for the view, regardless of where the data comes from, or what design patterns you use in your model for retrieving it.
You may have a bunch of layers of abstraction for your data loading operations within your model layer, but the controller should be calling some method, or methods, which at some point down the call chain, goes to whatever persistent datastore you are using and gets the data needed for your view.
The controller should be providing all of the objects that the view needs, as this is really one of its key responsibilities. Whether that means using the appropriate model object to grab the data from a database, or initializing a "view model" class to wrap all the objects and properties needed for the view to display, it doesn't matter.
Personally, I have always used Linq-to-SQL in conjunction with ASP.Net MVC, and have had great success using repository classes that grab the necessary objects from the data context. To allow my controllers to be unit tested, I use a dependency injection framework (StructureMap in my case) to have ASP.Net MVC provide my controller with a default instance of my repository interface.