Earlier quoted context omitted.
I do not see how it is confusing (to someone who is familiar with computers - i can see how some people who aren't used to computers might be confused by the maximization state). It is very useful in several types of applications, like image editors where you want to have multiple images open while having docked toolbars/controls, avoiding a waste of screen space and a clean background. This way you can have, e.g. th…
MDI's don't play well with multiple screens. While they have some good applications as you suggest, such as image editing - Spreadsheets and Word Processing were not great with MDI. The shared context such as pallets don't apply for documents.
About multiple screens you are right but a tabbed interface doesn't work with multiple screens either (while having all the other issues i mentioned) and having multiple toplevel windows means you get unnecessary duplication (and hence waste of screen space) of UI for each toplevel window, you have the window switcher (taskbar and alt+tab) "polluted" for each document view, etc which can be very annoying.
And really there is no rule against having multiple MDI host windows (outside of framework limitations - due to LCL inheriting the API from Delphi, the API assumes a single MDI form with multiple MDI clients, but the pure Win32 API doesn't have that limitation nor does Qt's own MDI implementation AFAIK) so you could simply have one of those per screen. Some tabbed interface programs do that (e.g. i think you can open multiple windows in Eclipse).
Though TBH personally i never use multiple screens so i don't know about that.