The overwhelming majority of retail games have been in 3D for a decade or more, and we've long since passed the point when the majority of indie games are 3D. Yet it nearly all standalone Python games are in 2D. Why is this? A couple of suggestions spring to mind - 3D is harder and more time consuming than 2D, for example, or that Python is too slow for 3D. But are these really the barrier that people have suggested?
Both of these rationalisations would seem to be pointing to a lack of suitable tools and frameworks for delivering 3D graphics in standalone Python games. For tools, there is of course Blender, which certainly has a steep learning curve but is at least sophisticated enough to be able to produce cutting-edge 3D graphics.
There are a variety of frameworks available, each with pros and cons. Alas many of these would appear to be poorly maintained and/or poorly documented.
More information on these will be forthcoming, but let's take a quick run-down of what's available:
Bindings for 3D Engines
There are a variety of bindings for fully-features scenegraphs/engines. These generally substitute a need to know low-level OpenGL calls with a steep learning curve of the classes and datastructures of the engine. They can require a lot of setup and compiling of dependencies, or a heavy SDK/runtime to download and install before a game is playable, which make it significantly harder for many users to run games using these frameworks. Some may not even be cross-platform at all.
Of the examples listed above, Panda3D treats Python as a first-class development language and as such is maintained and has extensive documentation, pyirrlicht is maintained but apparently undocumented, while Python-Ogre is less well maintained and poorly documented.
There are a few frameworks built from the ground-up for faster Python 3D. In theory, using Pyrex/Cython should mean that these frameworks are tolerably fast without being inconvenient for Python developers. In practice though, Soya3D is not infrequently maintained and has incomplete documentation, while PySoy is under active development, but unfinished and undocumented after 5 years of development (You are actively discouraged from using it yet - this is not esr's "Release early, release often" methodology). Both of these tie in a swathe of dependencies that offer far more functionality than just a scenegraph, at a cost of potential extra work to get up and running with them (Soya3D is broken on current Ubuntu, for example).
Pure Python Engines
Pure Python (based on PyOpenGL or Pyglet) have the lowest burden of dependencies, and is also the most legible and adaptable for Python programmers, but it is also likely to be slow - how slow, and whether that is a problem, is an open question. Each of these examples seems to be relatively poorly maintained, but nevertheless, they do indicate there is a body of useful Python code available to use. If you wanted to get a pure Python engine going, you do not have to start from scratch.
Have you used any 3D frameworks and how did you get on with them? Leave a note in the comments.