The sound quality of a VQF file is not better nor worse than a MP3 file, it is just different. Look at the pictures and you will understand what it means: when you encode music in mp3, the encoding process introduces some little compression artefacts. Instead of this, when you encode music in vqf, little details are lost and the sound is softened. So a 96Kbps VQF file seems to be more limpid than a 128Kbps MP3, but it is also less detailled. Two others problems of VQF are spatialisation (sound is far compared to the original) and pre-echo.
It is obvious than 96Kbps is superior than 96Kbps MP3, but it is not perfect nor really better than a 128Kbps MP3. But at very low bitrate (less than 25 kpbs) the VQF becomes really better than MP3. In fact, VQF would have been wonderful at 112 or 128 Kbps.
How does TwinVQ works?
TwinVQ (Transform-domain Weighted Interleave Vector Quantization) refers to a music compression technology that has been developed at the NTT Human Interface Laboratories in Japan.
TwinVQ is a transform coding method like MP3, AAC or Dolby AC-3. It uses some classical tools also used in MP3 (bitstream reservoir) or AAC (interframe backward prediction) but the encoding of music is totally different. In this method, the individual bits of music data are not directly encoded, but are combined into pattern segments (vectors). These patterns are compared against standard patterns which are prepared in advance. The standard pattern which provides the closest match is selected, and the number associated with that pattern is transmitted as the compression code. Data is packed into long frame mode or short frame mode (8 subframes) using a constant bitrate in order to enhance the error robustness. Coding distortion is minimized even at low bit rates, so music and other sounds are successfully regenerated that are highly faithful to their originals.
Notice that the TwinVQ technology will be one of the tools used in the upcoming MPEG-4 standard for natural audio.