Very apt description of the problem, I guess. Because when you render a complex sound, it might have sounded different on each keystroke, but the rendered version only has a single sample for any given velocity/pitch. So, it is essentially a 2d represention
There are a number of things you can do to improve the result. I would suggest rendering the sound not once, but twice or maybe even three times, as each render should result in slight variations in the sound (assuming that the plugin has internal modulation that produce these variations…)
Then you can use a tool like the xrni-merger to combine those renderings into a single instrument, and enable the overlap/round-robin feature in the keyzones to make the instrument become more alive.