SQL Server allows a foreign key to reference any column(s) that is guaranteed to be unique. The Books Online states "FOREIGN KEY constraints can reference only columns that are PRIMARY KEY or UNIQUE constraints in the referenced table or columns referenced in a UNIQUE INDEX on the referenced table."

When the foreign key is created, SQL Server will bind the foreign key to the index. There is no documented behavior on which constraint or index will be chosen when multiple candidate unique indexes exist. The rules SQL Server uses are apparently different between SQL 2005 and SQL 2008+. I don't think it's correct to classify a change in behavior as a defect unless it's contrary to documentation.

With some experimentation, I learned that SQL 2005 chooses the clustered index on a table and the first non-clustered (lowest index_id) on a heap. However, SQL 2008 and SQL 2008 R2 choose the first non-clustered index on both tables and heaps; the clustered index is used only if no candidate non-clustered indexes exist.