Some links on this topic. All based on blind tests.
December 2005, the last multiformat public listening test at 128 kbps (actually around 140) : Results of the public multiformat listening test @ 128 kbps (December 2005)
All of the 18 samples were rated above 4.0 with all 4 encoders, except maybe one, encoded with mp3, on the following scale :
1 - very annoying
2 - Annoying
3 - Slightly annoying
4 - Perceptible, but not annoying
5 - Imperceptible
Which means that 128/140 kbps compression was barely perceptible for the group, made of about 30 listeners.
June 2006 : the ABX Just Destroyed My Ego - Hydrogenaudio Forums
thread on hydrogenaudio.
All of sudden, forumers started ABXing their own encoder. The thread starter could not hear the difference between 64 kbps and the original, while he was, before the test, reluctant to use lossy compression, even at 320 kbps, instead of FLAC, and could not imagine listening to something as a 128 kbps mp3.
Many listeners could not distinguish between 128 kbps and original.
Soundexpert tests with artifacts amplification : Audio quality of encoders at 128 kbit/s - SoundExpert
Many codecs are rated above 5 in the 128 kbps test, which means, according to their documentation
that their distortion can be considered to be outside the hearing ability of the listeners. They used special tricks in order to amplify the distortion on the samples under test.
However, these are collective tests, made with many untrained listeners. On the other hand, some tests made by trained listeners show that at 175 kbps, for example, differences are still audible, even without using especially difficult samples : MPC vs OGG VORBIS vs MP3 at 175 kbps - Hydrogenaudio Forums