Like others have said, I agree that a complete game engine as one boost library would be (1) too much of a mix of everything and (2) too domain-specific for the "general-purpose" vocation of Boost. Also, I don't think that the Boost community has any intention of starting to provide libraries with graphic displays (GUI or 2D/3D graphics), which is a domain that is huge and best left to other dedicated projects (IMHO), like Qt, SDL, Sfml, OpenGL, VTK, OpenCV, etc.... However, I have an idea of one related thing that you could consider as a GSoC project. You could write a generic resource managing library, which is the kind of thing that game engines have for files, models, sound tracks, textures, etc..., where the central resource manager takes care of avoiding re-loading the same resource twice, unloads resources that are no longer need (on a per-level or per-scene basis). This is general-purpose enough for Boost, and is one of the central component of any game engine, and is a more manageable project to do in one summer. However, writing a very nice generic resource manager could be a hard task that requires very good skill in C++ generic programming.
We are still finalising the rules, but if you have previously written a 5,000-10,000 line open source C++ library that would definitely count.
Make that 20k to 50k lines ;) I'm happy to see some kind of minimum bar because GSoC has been abused a lot by people who really lack the experience to write code of sufficient quality. Cheers, Mikael.