I noticed that with lots of the Autocad projects I’ve seen, if you have an external plan file that you want to xref into a layout file with a title block, the normal practice seems to be to xref the plan first into model space, and then place a viewport of it into a paper space layout.  
Why isn’t the best practice instead to just xref it directly into the paper space layout and avoid the viewport step? (This is if all you’re adding is notes and dimensions to the plan and no new linework.)

`You could do that, but then you run into a scaling issue. where x-reffing into model then creating a viewport allows you to select a scalable view that engineers and such can measure the drawing appropriately.  now you can x-ref a title block directly into your layout.  I hope this helps explain this. 

Hi. Generally a model is drawn in modelspace (or in this case XREFd into a modelspace). Usually, it is whatever is in modelspace that gets XREFd into another drawing, and you may want XREF something from drawing A into drawing B and then XREF the entire thing into drawing C for example.
 
To transfer an XREF into paperspace kind of defeats the whole scaling thing, in that you would have to clip you XREF or scale up your border (and scale down your plot output). You may even end up needing multiple copies of the XREF to rescale for different tasks, but then there's the whole layer issue to deal with. If you wanted to display the XREF twice (once for layout and once for RCP for example), you'd need to rename the XREF and then insert a new one (not just copy), just to be able to turn some layers off for one and different ones off for the other. Then the layer list grows huge.
 
These days, at least in AutoCAD, one XREF does it all (in modelspace). both modelspace and paperspace are at a 1:1 ratio, so in paperspace you define the size of your output sheet and save the parameters (so you don't have to keep defining what to plot). Then you can arrange different views of your model on the paperspace "sheet". They can be at different scales from each other. They can have different layers frozen from each other. And nowadays with AutoCAD you can override the linetype and color of each layer in one view, while keeping it original in the other one. There's even an "annotative" feature for dimensions and other objects, so one size of dimension fits all viewport scales.
 
That was an oversimplification, but I hope it helped