Dear all,
Thank you for your lengthy responses, things seem to be much clearer in my head now.
By reading the replies and ff123's article I think realise three points: a) Clipping is an inherent feature of the mp3 decoding-encoding process as related to the necessary time domain (integers) <--> frequency domain (floating point) transformations. b) Clipping during the reproduction of digital audio is also inevitable, due to sampling effects, because of players (domestic or computers) trying to restore the waveform parts contained within the samples. c) If I understood well, these two effects are interrelated. (a) implies that any wav file close to 100% risks of being clipped, and (b) implies that the extent of clipping depends on the type of music program encoded. (b) also for me implies that whenever there is high frequency content present in the wav (e.g. cymbals), the decoding process has few samples to process and moreover the respective analog waveform has acute peaks, so repeated short-term clippings are inevitable. On the other hand, low frequencies (violins, voice) have smooth peaks so when when clipping occurs, we have a long-term clipping i.e. a flattening of the waveform.
I am convinced that both of these phenomena, detected by the peak detector of Mad, are quite audible - perhaps the low-frequency cases are a bit easier to detect because of higher ear and brain sensitivity to them. I wonder whether modern recording engineers are fully aware of these effects. The "Remastered" stickers on all those new generation CD's make me think ...