The linked article doesn't mention call stacks explicitly, but describes the R1000 arch was object+offset addressed in HW. So unless they restricted the call stack to fit into one object and use only offsets, then yes, they must have chained objects together for the stack.

When you have a page-based memory model, you've created the importance of address locality. If you have object-based memory model, and the working set is of objects, not pages, then address locality between objects doesn't matter.

Of course, page-based based memory models are by FAR the most common in practice.

(Note: pages ARE objects, but the objects are significant to the VM system and not to your program. So strictly, page-based models are a corner case of object-based models, where the objects are obscure.)