yeah but what if in the voxel room you decided to change the face proportions tremendously, and sculpt the ear a lot to find different shapes (not only by moving around and posing, which works in both retopo and voxel rooms), taking into account how delicate the ears topology is, you would then have to adapt manually the retopo mesh into the new voxel mesh cause just snapping it would make a mess.
Besides, what michalis says is that this subdivision adaptation works great on some cases, and can even be more useful than voxel-retopology workflow, depending on the case, and it would be useful to have the chance to do it that way. On the other hand, subdivision baked displacement maps are much more trustworthy than raytraced ones, at least that is what happens in mudbox. I agree with michalis in that is generally easier to work over a mesh with good topology already, but the downside of that is that you can't skecth and design a character without having anything in mind, which is in my opinion one of the things 3d-coat shines at... and also, another downside of subdivision sculpting is that you can come short of polygons if you want to add, for example, a horn in the head, and you may get to insane cuantities of memory usage, and that horn may have troubles with higher subdivision levels if it's added by booleans (I don't know very well how boleans work over meshes with subdiv levels in zbrush though). Even with this downsides, there are many upsides that have been posted along this thread, and it would be more than helpful to have the chance of doing it the subdivision way, like when the mesh has many folds and tiny holes that is a nightmare to retopology and bake maps with raytracing.