Quote - Just for those who have difficulty understanding: A cage is a 'mesh' of which the verrtices (usually called 'nodes') are attraction points, not surface definition points. Compare it with the positions of magnets when you try to pull a flat surface into a shape using a grid of magnets covering a certain local part of the mesh. So, if you want to model a local hump, you have to pull the magnet governing the spot far out, so, if you represent the magnet positions as a mesh, it will show a local peak. 

We are therefore looking not at a geometry definition, but at raw data that is intended to be used in a specific process to make a geometric definition. That this raw data very much looks like a geometric definition and in face can be represented as one if you take the node positions as vertex positions and take the cells between the nodes as polygons, can be convenient in some cases. The cage, so represented, even looks very much like the final geometry. 

Catmull-Clark subdivision (or rather interpretation of a mesh as a cage and generation of a new geometry from this cage) causes drift of the definition. The new geometry is not where the original vertices were. Look at the famous example of a cube transforming into a sphere:(http://en.wikipedia.org/wiki/Catmull%E2%80%93Clark_subdivision_surface) The corner points of the original box are not on the surface of the sphere. There are ways to fix this, of course, but those are to my knowledge not applied in DS.

For that reason this type of smoothing can not generally be used for any purpose where accurate interplation of the surface trough the definition points (which may be results of measurements) is required.

The drift of the definition (or the distance between the smoothed surface and the original definition point) is proportional to the ratio of the distance between the nodes of the cage and the average radius of curvature. A ring defined with a coarse cage may  disappear into a finger defined by a fine cage.  

Also note that when a surface is subdivided it loses detail. The only information used is the XYZ position of the nodes. Even the information in the normals is not used.  Depending on the distribution of distances and angles between the 'edges' of the definition grid, local exitation can occur and, yes, when used in a cunning way, this can be used to some extent to simulate detail. 

I think that most of the misunderstanding comes from the fact that the "cage" is the thing being exported to poser. If it had been een single subdivided mesh, it would have looked better and this misunderstanding would not have happened.
If they have done this deliberately, I don't understand the rationale behind it and the instruction to specifically choose the "high definition" mesh is a complete mystery if the base mesh/cage is exported

I think most of the poser users who were interested in genesis gave up after they saw the ears