Let me introduce a relatively simple real-file example: in the Shelley-Godwin Archive we are working with 3 volumes of Mary Shelley's Frankenstein as labelled by the Bodleian Library.
The shelfmarks correspond to three boxes of loose sheets that used to be part of 3 bound notebooks. The notebooks did not correspond one-to-one to the volumes (e.g. Notebook A contained Volume I and part of Volume II).
When we encode the *notebooks* with mdDesc, each msPart needs to identifies a volume (or part of a volume) as the part of a former manuscript.
With this shift, I don't think there would be any change needed in the content model of msPart, so I would be in favour of extending the definition of msPart and introduce values for @type to distinguish usage.
One way of not breaking backward compatibility could be to consider a non-typed msPart to comply to the current definition.