open source and getting paid are not mutually exclusive.
e.g. several Linux developers are paid by their employers, like IBM and HP and Motorola and whatnot, for their work. also Apple releases their Darwin kernel as open source, and has actually hired FreeBSD core people to work on it. and Safari uses KDE Konqueror KHTML derived code, and Apple’s improvements are put back to the open version (also Nokia uses KHTML nowadays somewhere iirc).
i personally am paid to work on Blender right now. also much of 2.40 was paid work, by Google via the Summer of Code program. but very very much was not, but just done by people because they want and need, and that is great and i definitely dont want to put that down with all this talk about ppl getting paid for open source development!
if there are people willing to pay to get features in Blender, the open source community has established mechanisms for that. like Gnome has a bounty program i think, dunno how that has worked (i dont use gnome nowadays). oh and KDE KOffice or something has now a UI design contest with a 1000 dollar price or so.
i dont know if this has been discussed w.r.t Blender Foundation but it certainly can be at least discussed (sorry for the stoopid sentences am really tired now)
back to the actual topic, the CrystalBlend integration would be that Crystal Space would be bundled in the Blender release, and integrated so that just by pressing ‘p’ it would run the GE, in the optimal case just like Ketsji is run now. so users would not even have to know it is crystal space, it would just be the Blender GE. and same can be done with Ogre and other graphics renderers if someone prefers them, the CrystalBlend code can be reused in that too (as also Ogre is C++). but this is not very tight integration with Blender in the sense that much of the code to drive the animations etc. would be separate, i guess pretty much like with the current engine.
but your original question is interesting. i did mention that one other old engine, Enji? it was more integrated in Blender than the current Ketsji, and used the same code as Blender itself and ran in Blender afaik (have never seen it, just read once what Ton wrote about it and talked about it with him once, and i was impressed!). if Enji was still there, perhaps even particles might just work there … i dont know. but Enji was really limited, not as much powerful as a GE as Ketsji, but i guess that could have been developed … and perhaps that development would have benefitted the 3d views of the animation/modelling side too? where we need improvements … like getting opengl2 shaders to work etc.
one reason for modelling/animation and GE being separate is the different demands for timing the execution. i mean, in a game you usually want a constant framerate. and usually dont need complex mesh manipulation like extruding capabilities, so can optimize the running. in modelling / animation, you want to see the thing exactly as it is, and have all the complex tools readily available. as you probably know with a big scene and complex operations (like fractal subdivide or whatnot) blender can get really slow, like 1fps, even on a powerful machine. that is the desired behaviour when modelling / animating, but not in a game. so the core of the code and how all the operations are written are different. this is one of the things that makes seamless integration of modelling/animation and gaming … challenging at least, and certainly impossible in the minds of many, and in near future in any case. but i dont think it is absolutely impossible.
i hope i managed to clear the issue up a bit.
thank you for the enthusiasm!