MultitrackStudio's internal data format is 32 bit fixed point. Every time truncation to 16 or 24 bit is performed (for instance: when sending the mixer output to the sound device, or when saving files created by the Sample Rate Converter) adequate dithering is applied. Dithering turns distortion (caused by truncation) into random noise. This noise is less objectionable because of its random nature. The dither signal used is called HP-TPD (High Pass Triangular Probability Dither).
Noise Shaping can be used to move the dither noise to frequencies the ear can't hear very well, thereby improving the perceived sound quality even more. However, this type of dither should only be used when you are sure the file will never be processed again, as doing so will increase noise. MultitrackStudio uses third order noise shaping at 44.1/48 kHz, and second order noiseshaping at higher sample rates. Noise Shaping is applied only when truncating to 16 bit, not when truncating to 24 bit.