The problem is that Aggregations and Compositions cannot be distinguished from regular associations in code because the .NET languages (apart from Visual C++ perhaps) do not distinguish members held "by reference" (regular association) from those held "by value" (aggregation or composition).
Even if you could model aggregation in Class Designer, the information would be lost when you re-visualized the code, for example by dragging the classes onto another class diagram.
This is where the VS2005 Class Designer differs from UML class designers. The VS2005 is a 100% visualization of the code, with no translation. In UML tools you have to "code generate" and "reverse engineer" between the code and a model that holds additional metadata (such as whether a particular association is an aggregation).
Tony Loton
Co-author, Professional Visual Studio 2005 Team System, http://www.amazon.com/gp/product/0764584367