Because encoding to mp3 and then decoding again is a lossy process, clipping can occur. In the case that the peaks of the source are already at 0dBFS, clipping is practically guaranteed, especially with titles that are compressed in order to make them louder. IOW, even with simple normalization prior to encoding to mp3 you are likely to encounter clipping when decoding.
The idea that mp3gain does not use ReplayGain is simply wrong, too. How else would it know how much to adjust the level of your mp3 files assuming you tell it to do so? Furthermore, by default mp3gain writes RG information to a tag whether or not to tell it to adjust the level of your mp3 files. If you're unable to make use of the RG information in an RG-capable player it is because the information is written to an APE tag rather than to an ID3 tag. Copying it over to an ID3 tag is fairly straight-forward.
As David suggested, you should have a look at the manual. It would probably be helpful to have a look at what ReplayGain actually is and how it works as well since you're wondering about the reference (which is not 100dB, BTW). While mp3gain is capable of peak normalization (or close to it, since adjustments can only be made in 1.5dB steps), ReplayGain was intended to be a superior alternative to peak normalization. In many cases it actually reduces the volume or your tracks.
http://replaygain.org