That's well said, you nailed the pros and cons of sample technology. Like violin, especially saxophone is another very problematic example. The sound is so much influenced by the player's flow of breath, lip movements etc. and changes throughout one note, so many ways to bind two notes, these things will probably always sound too static when played with a keyboard. Maybe a very advanced physical modeling would be able to come close in combination with the use of a very sophisticated kind of midi breath controller.

Roland's attempts to gain back the subtle control by implementing SuperNatural behaviour modeling sounds have not been too convincing IMO. Instead, regarding some sounds they have even lost some realism previously achieved with some of the SRX expansions. Behavior modeling is only convincing if the "model" used is much more complex and realistic than what Roland has developed so far; I guess they should go back and implement better samples and work them into the modeling technology (advanced sample switching etc).