Hi Helmut,
This is true, however it's a bit more experimental than desired; what if you are not prepared for an unforeseen scenarion in the code? But actually, the bigger issue is that I'd be particularly interested in configuring this for transforming a domain model to a C# model. To conform to RUP and other principles, in the domain model you should not concern with the data structure used to store a collection of values; it is rather okay to call it array. However the heart of implementation design and the supporting C# model transformation is to transform UML data types/syntax to C#.
To response to your suggestion more precisely: EA will define the return type as List<Foo> when reverse engineering a method with List<Foo> return type. However in case of List<Foo> return type in EA, EA will not recognize Foo is part of the semantics of the model, so if you rename your Foo type in your model, List<Foo> will remain intact as a return type (but if it were only Foo, it would be renamed to the new name). Similarly, when generating code from a type like List<Foo>, EA won't know what namespace to specify for Foo, 'cuse it will only be a char literal for EA.
So I'm afraid checking the results of reverse engineering does not answer the conceptual part of the question, and yields in a data type EA will not handle as expected.