VBR streams frequencies deemed necessary for efficiency purposes.
So, songs that contain several instruments playing all at once will naturally output a higher bit-rate. The reason on why it's regarded as more "efficient" is because bit-rates (while the song is playing, decoding) instantaneously adapts by streaming JUST the right amount of memory necessary for instrumental playback. So, when there are NO frequencies, naturally you have NO memory streaming.
Basically... when you have (for example) 600 frequencies within a frame, VBR exerts X amount of memory to accomodate those 600 frequencies; no more, no less.
CBR is exactly as it sounds -- CONSTANT. Constantly streaming any set bit-rate for the duration of any given song, from start to finish.
So, even when you have no frequencies, NO sound -- you still have that set bit-rate streaming as if there were sound to be played. PCM/WAV uses this.
ABR (in essence) works the same as VBR. Only difference here, is they just went ahead and added, then divided for a mean avg. rather than streaming each frame live -- as the song is being played back. FLAC uses this.