Why Noise reduction is a DSP function that tries to separate speech-like or steady tone components from random noise. As you turn the NR level up, the algorithm gets more aggressive and starts removing parts of the wanted signal along with the noise, producing hollow, warbling, or watery-sounding audio. That processing artifact is heard as distortion of the received signal, so NR is usually set just high enough to help readability and no higher.
Watch out The idea that CW gets severely attenuated is tempting because heavy NR does make CW sound odd, but NR is designed to preserve tones and it does not knock the signal down; nothing in the audio DSP chain changes the receiver's tuned frequency or oscillator stability either.
NR works after the mixers, on audio only. It can never move or destabilize your frequency, but crank it and audio gets mangled.