Actually I don't understand why it's so important that the 3DMark engine must be used in some games to be qualified for a "game benchmark."
Computer games have very high diversity. An engine for one game can have very different performance characteristic from another game engine. Therefore, there is no so-called "representative" game benchmark. There will never be such a thing.
I think, what 3DMark wants to be, is a kind of prediction of how future game engines may perform. Of course, no one knows what direction future game engines may follow. The folks at Futuremark are just guessing, as everyone else. As long as the engine in 3DMark performs similar tasks as a game engine, I think it'll be a good estimation of how game engines may perform in the future. Of course, there's something quite important in a general game engine design that 3DMark can't ignore. For example, in 3DMark, the camera normally follows a pre-defined path, but this is very rare in an actual game. An engine can be designed for a fixed camera path, while a game engine can't. So, the engine in 3DMark shouldn't be optimized this way, or it won't be a good estimation. I think the "game mode" in 3DMark is a good demonstration that the engine is not designed just for a fixed camera path, among other things.
Maybe someone can do a nice review over the last few 3DMarks in retrospective? Like, seeing how "vintage" graphics cards perform on games released years later than a 3DMark version, and compare how well the 3DMark estimations are?
