.OBJ is still good because, as you say, its a good interchange format.
If static models could be .OBJ then it would be that much easier for others to edit; no chasing around looking for an original nor needing an importer.
I don't actually see that helping at all, since a very small number of models are static (usually just tileset and some buildings), and having to use two different importers for inconsistent model types would be a hindrance.
MD5 is bone-based and, if Glest supported it, would be super. MD5 would need some special bone naming convention such as 'turret' and 'barrel' and flags in the name string so we can know which bits to tilt with the terrain and to face the enemy and so on.
MD5 would definitely be an improvement, on the other hand, as it supports animation, could cut back on filesize (since you'd have one file that describes the mesh, and then numerous files for the bones, rather than having the mesh and bones for each animation), could potentially be used to allow turrets as Will mentioned (by naming specific bones (or meshes?), with projectiles that have an offset, perhaps you could refer to a bone name for the location of this offset, could allow animations to be "merged" rather than replacing the model with a new one (provided the mesh is the same), etc.
As well, the importing and exporting of MD5 is far, far better than G3D. Rather than our homebrew importer and exporter, there's existing importers and exporters for various modeling programs, so we wouldn't force people to use Blender. Plus, you could import animations (the inability to do so with G3D is a royal pain). In addition, it's a better known format, so a knowledgeable modeler won't necessarily have to install any extra software, but could likely export to it easily enough, and the format is well documented.
Original GAE discussion on MD5
