First off, I just want to point out that the code is still a work in progress, and shouldn't be considered representative of what to expect yet. That said, the normalmaps for the geoscape and the normalmaps for the battlescape are not actually handled the same way. The reason for this is that the geoscape uses orthographic projection, and the battlescape uses perspective projection. The alpha channel of the geoscape normal maps is actually used to specify specular reflectiveness; the alpha channel of the battlescape normal maps is used to specify height (for parallax, which is meaningless for orthographic projections).
There are also some issues with the models; all the normal vectors for the verticies of the earth on the geoscape are correct, and nicely smoothed. It's easy to get them right, because it's a sphere. For the arbitrary meshes in the battlescape, there are places where two triangles share a vertex (in the actual .md2 file) but where smoothed normals look very strange. This results in surfaces where the mesh looks flat but the normals look curved; while this is nice for smoothing things that *should* be curved, it looks strange when applied to sharp corners. I'm planning on writing some code to attempt to de-construct and re-construct models to fix this, but haven't done so yet.
Another issue is that the light sources are generally not in locations that make any sense. This is because up until now they never had any effect on the actual appearance of things. The lighting system is going to need some work before things will really start to look good.
Finally, load-times in the latest build are long; this should be fixed soon. Right now all the normal and tangent vectors are being calculated at runtime, but Mattn is working on a tool to compute them off-line and store them in data files. Once he's done, load times should go back to the way they were before.