"Compared to an acoustic harmonica, the DM48 is less expressive in many ways, as mouth geometry, hands, and classical bending/overblows cannot be used to modulate the sound. [...] the DM48 cannot in any way replace or convincingly emulate an acoustic harmonica."
It could be that it's not the input latency itself, but rather the tapered nature of the wind instrument attack curve. If a tapered attack curve is used to set of the playback of a sample, that could in itself also introduce extra input latency. It's not entirely clear what's happening here.
The comparison is like the difference between driving a truck and a car compared to driving a car with a 500ms delay on its inputs.
Same thing happens when you play an instrument.