What if you process a V3 uv template and compare the result with Antonia's? That should give you an idea of how to solve.

When there is a shift of the uv islands to one side there may be a rounding off error.  

When the uv islands are a tad too small some logic is needed to grow the uv islands a bit.

Allow me to give some idea for this: A seam means a 3D vertex is represented by more than one vertex in 2D, so that's a way to identify uv island edges.

Orientation of vertices around a facet follows right-hand rule so in the morph target an edge leading from one edge vertex of a facet to another needs to be shifted perpendicular to that edge, negative normal direction. To avoid doubles this shift should not be done for the last edge vertex of the facet because that one will have it's shift with the adjacent facet.