So, I moseyed on over to The Wikipedia page for game engines (http://en.wikipedia.org/wiki/L.....me_engines
) and there are no less than 59 free or open-source systems listed that are implemented in everything from C/C++, through Java, Python, and Pascal, to PHP, and everything in-between (I"m sure someone has a FORTRAN version running in their basement! ). As police chief Brody would say, "You"re gonna need a bigger boat." Holy Moly! If we limit ourselves to a Python implementation (only because the Foundation is leaning in that direction), the possibilities drop to five: Blender, Cocos2D, Ignifuga Game Engine, Multiverse Platform, and Panda3D.
The Blender engine appears to be built into the modeler tool, so, I assume it"s just a component that can be run in the background, but, if the entire system has to be run, that may not fit in a Pi"s limited RAM, especially since half the RAM has to be dedicated to the GPU at boot time if it"s going to be used, leaving around 80 MB for user applications and services such as Blender. On my Debian x86 system, Blender is using 118 MB, so, that could knock it out of contention if it needs a similar amount of RAM on the R-Pi. Virtual memory might be workable if only small sections of code really need to be in RAM to keep things running - really wish I had some hardware to run it on ...
Cocos2D and Infugia are 2-D only, so, those are non-starters, and Infugia is only an alpha version, anyway. Multiverse (MIT) appears to meet the desired requirements of Pi-finity! the closest, based solely on the brief features summary on the Wikipedia page and, although they ceased operations in December 2011 due to funding shortfalls, the code and documentation is available while they reorganize financially. Panda3D (Disney) Is 3-D, implemented in C++, has C++ and auto-generated Python interfaces, and can import models from Blender, 3ds Max, or Maya. Evaluation of these will keep me busy for a few days, and then I"ll be looking at links to games/tools that others have provided here and elsewhere.
Early indications are that we should at least be able to run MMORPG clients on the R-Pi with traditional servers doing the heavy lifting running game engines. I may pursue the P2P possibility experiment once we have a stable baseline server-based system running with a reasonable number of users and we have had time to determine how to best use the R-Pi boards, taking into account RAM needs, virtual memory usage/performance (most likely via an external USB hard drive), network loading and performance, etc. This is very much in keeping with the Pi-finity! Prime Directives of leveraging existing code as much as possible, building the system via modules with well-defined APIs, using rapid prototyping to try out new ideas (in parallel where feasible), and documenting everything in the wiki.