The real problem in my experience is the ability to play rather loudly with very low distortion and flat frequency response. As you probably know, the way the ear-brain system works, reproduced sound will only be perceived as accurate if it is played at the same loudness as the original. Flutes are relatively easy in this regard. (I had my step daughter once play along with a recording of herself, standing between my speakers, and it was difficult to tell the reproduction from the original.) It takes a hell of speaker speaker to fool someone with an upright piano, and I haven't heard it done with a 9' grand piano. My Salon2/Velodyne DD18plus system in our 9000 CF+ living room can't fool me with my wife's DW Jazz series drum kit, which is pretty modest in size, though if she is not rambunctious the reproduction is very good. Her rock drum kit, which has a 22" kick drum and bigger toms, is untouchable by my system. The reproduced version is blatantly obvious.
A marimba is louder than one would think, and my wife's is only a 4.3 octave version, not the in-vogue 5 octave versions like orchestras use. The system does a good job, but doesn't fool me.
I admit to being attracted to large full-range electrostatics, like the Sound Labs or the Martin Logan CLS, though they definitely need subs for most music. They sound "real" in a way that conventional speakers don't. Have you ever heard one?