The underlying theme here is if organization of knowledge in a network or in a hierarchy is better. In my opinion this is the wrong approach to think about it. These two structures are not mutually exclusive, they can coexist. Hierarchies are effective for large-scale, slow-moving efforts while networks are good in small-scale, quickly changing situations.
Dynalists approach is to apply a network (with tags and links) on an underlying hierarchy while roam does the opposite. I personally think that the first approach makes more sense and can be applied to more use cases.
The drawback people see in the Dynalist approach is (like you wrote too) that they need to re-organize more in Dynalist. I think this is a sign that the system people are using is not optimal, not the tool. With the right system you need just one shortcut (CTRL+Shift+M) to move everything to the right place. Keywords here are P.A.R.A or the Zettelkasten method.
The lack of a real hierarchy in Roam makes it easier to use not necessarily better in the long run.