i would assume, that you _are_ triggering a hardware issue. your description sounds like your hardware is running with 16 bit depth. at some point 32float values are converted to the resolution of the hardware. in case your hardware really is running only 16bit, then it is very likely that you hear some artefacts on those very low levels. what you hear then is the quantization error, which is quite perceivable at 16bit,
Oh yes of course!!!!! :$:$ You're definitely right.
With so much talk about float precision in the cpu and interpolation and stuff, I forgot about the final quantization of the soundcard :$:$
I didn't know it was _that_ audible at 16 bit however.