Shared vertices or vertices for each face?

Hi folks, considering procedurally generating an object: I see lots and lots of examples for generating objects using either shared vertices or (vertices = faces.count * 3) style programming. As far as I can tell, these are the differences:

Shared vertices:

1: Are much harder to UV map (can be impossible, depending on what you're trying to do)

2: Uses less data and can be faster

Vertices for each face:

1: Number of vertices is always number of faces times 3

2: More data to manage and work with

3: Much easier to UV map

The real questions I have are:

1. Does soft shading work with vertices for each face? It seems like it wouldn't, since it cannot see the normal of all adjacent faces (i.e. each vertex is attached to only one face). How does that work?

2. Does collision (using colliders on the mesh) have any preference in unity? Is one faster/slower? Is it prone to error to have a different set of vertices for each face?

Thanks!