I've read several of the posts on normalizing audio and have found that it is a good idea to do so but one question remains that I was unable to find a clear answer for (which there might not be), which is better to normalize to, the PEAK or RMS value?

I've seen some WAV files where even if you normailze the PEAK value to 0db the overall RMS value doesn't change by much. Also, if you have 2 audio files but with different peak values, if you normailize both PEAK values to 0db, the 2 files may still wind up sounding very different in overall volume. Would this make it more of a reason to just normailze the RMS value? With using RMS value to normalize some bring up the issue that this can cause clipping. Even if dynamic compression is used to prevent clipping, wouldn't this alter the way the audio was mean to be heard as far as the depth and pitch? Any experts out there that have experience with this and can suggest the best way to normalize. How do they go about doing this in holywood?