I am not quite sure about this but here goes
When most people encode to mp3 they use mp3gain to make all their mp3's the same loudness
But of course mp3gain is lossless but if you scale the sample in lame before you encode it is lossy
But here is what i understand if you encode to mp3 you most certainly get clipping and hence when using mp3gain it will lower the volume of your songs to the same volume so your music has the same loundness, and winamp etc... will not clip badly when playing the song back.
But when lame was encoding the song it clipped therefore loosing data and getting distortion and when mp3gain lowers the volume it still has the data lost but looks like it hasn't clipped when viewing the waveform but still has the sound of a clipped sample.
So even if you scale samples it might be lossy so why
I tell you why, say when encoding with lame you use --scale 0.9 or something similiar it will stop lame from clipping on the song, therefore no distortion and then you can perform mp3gain to achieve the same volume.
As now it is late and i am tired, so i might be talking complete crap
