Unless the recorded level of a single track is too low to use in a mix situation, I don't see any advantage of normalization - and every pass of it degrades the original signal, even though it is digital. It works based on peaks, so also for mastering the use is very limited and can get you in trouble easier when further working with the master (conversion to mp3 for instance). The usual recommendation for masters is to stay below -0.5 to -1 dB True Peak level.