> I'm interested to learn if those unsupported features are potentially useful
It depends entirely on what your data represents and how you intend it to be used.
Inheritance would be useful if you want to be able to define a general concept like e.g. Furniture plus various subtypes like Tables and Chairs and be able to query for Furniture and get back both Tables and Chairs. It requires users to understand the concept of inheritance and requires the application to avoid the many pitfalls thereof (multiple inheritance, name collisions, etc).
Calculated properties are obviously useful and I hope to see them become available to Item Types. The user interface would be an interesting challenge, which we've already tackled to an extent in the Variables dialog.
Read-only properties are not obviously useful if the user is creating the data directly (why would she restrict what she's allowed to do with her own data?) Use cases do exist, e.g. the concept of 'scope' in the Variables dialog.
Nested structs are useful if you've got compound data types which contain other compound data types.
Schema references are useful if you want to define some reusable types like maybe an Address type in on Item Type Library and then make use of that type in other Item Type Libraries, but now you have a dependency between two or more libraries so workflows which want to import or edit libraries as discrete entities will no longer work.
Most of the other unlisted unsupported features are only useful if they are backed by application code defining the behavior of the data.
Also just a note/warning regarding this:
> export the schema and open it in Bentley Class Editor
Don't edit an Item Type Library schema in Class Editor and then try to use it in MicroStation (or, if you do, don't complain when things break).