A giant battle might be best to test overall FS2 performance, but I am somewhat concerned that a Battle of Endor mission would not provide repeatable results. The AI decisions will always be somewhat random. Only way to determine how this will skew the results is have the mission and try it numerous times. Oh, the player ship better be cloaked, disabled, and not too close to the action. Else, it will be an awfully short test.
An idea for a mission that should eliminate the random AI ‘factor’: At regular intervals, say once every ten seconds, a large battle group warps in. Each group follows a waypoint for about twenty seconds, then warps out. No ‘dancing’ or colliding caps, please. At any point in time, there would be two battle groups on screen. All ships are friendly. Of course, without weapon fire, there is no way to determine performance impact of weapon handling code changes, special weapon effects, etc. Still, it would provide a good gauge of basic engine performance (lots of polygons and textures to render, and the warp in/out exercises the ani code).
Maybe one of each?
Oh, important: Either mission should end at a set time (is there a mission-end SEXP?). Again, repeatability is the most important thing here.
To get a meaningful benchmark, it will be necessary to add a little code that calculates the average framerate for the entire mission. Easy to do (most of my real-world software/hardware designs include data acquisition and signal processing).