I mean that I want to draw controls using true depth feeling, which could be made only with projection matrix. With ortho I need to scale objects manually if I want make them smaller when they are far from the camera.
You will want to just do normal 3D rendering and use a camera to view the menu through. Use this camera idea to frame the UI too it makes it a lot easier to deal with.
In a 3D ui you dont have to care about the 0,0,0 mapping on to the 0,0 corner, this is what the frameing with a camera is for.
If you look at the DiRT and GRID games they all have a 3D ui, and for ease of development we never supported mouse input all input is either keyboard arrows and enter or controller. Doing the unprojection is not that hard but the system we had to create these UIs made that a lot harder, the update side of the UI had no idea of where the rendering of the UI was happening in space. Which allowed for a lot of flexibility in setting up scenes and implementing the behaviour of the screens at the cost of not having mouse input though.