I am quite familiar with mp3gain, it works fine for me. 97% of 2500 albums do fit to 89 dB, sometimes it is necessary to go down (as far as to 83 dB) when peaks are too high (namely on albums from "ages before compression"). No problem, when your device is not poor.
Now I need to help to my friend with the same thing - but with his ogg archive. I found vorbisgain set up to 89 dB and I do not see in it any switch which would allow me to set it down.

I consider to be important this: how does player (winamp + in-vorbis) handle the combination of parameters in tag (REPLAYGAIN_TRACK_GAIN, REPLAYGAIN_TRACK_PEAK) in those 3% of cases? (Each album in one ogg, that's why I am going to use track-gain.)
- output is set accordingly to track-gain, peaks if higher are simply cut
- output is set accordingly to peak (peak normalization) and gain is good for nothing
Which one is right?

Let's talk about this (yes, extreme) example: REPLAYGAIN_TRACK_GAIN = +4,80 dB, REPLAYGAIN_TRACK_PEAK = 0.99781585 (that's -0,1 dB I guess).
Do I have to manually change gain in tag (this time to zero) in such cases?

Thanks for your help.

I don't know winamp but all I use have option to limit volume raise to avoid clipping.

The 89dB in replay gain, as far as I know, is just the loudness level used to match the level against pink noise. The target level is -20dBFS I think.

I also don't know Winamp either, but if the player or plugin has followed the instructions for how to implement replaygain, then there should be two different clipping prevention techniques to choose from; a limiter like e.g. a -6dB hard-limiter or an advanced read-ahead type limiter and auto-attenuation. The first option compresses the peaks if they would clip otherwise and the other option attenuates the whole signal if there would otherwise be clipped peaks. The first option will remain the replaygain suggested loudness during clipping prevention, but introduce compression on the audio, and the other option will not compress anything during clipping prevention, but the loudness will be lower than the replaygain suggested level.

Thanks, Martin cleared the situation up. No matter that I did not catch, how these two techniqes are choosen, because both are bad for me:
- either the raplaygaining concept is corrupted (another level is used in contrast to other reaplaygained tracks)
- or the peaks are compressed (we love to listen to all peaks, don't we?)

That's why I guess I was right, when I was affraid of manual changing the gain in tags - because vorbisgain does not know this. (Fortunately, they are those 3%.) Any disagreement?