This is bunk. Its supporters have never offered properly controlled experimental results to back up their claims, and have consistently failed double-blind tests. Quantitative studies of the digital data coming from the digital portion of the CD reader (before it is converted to an analog signal) show that it is 100% identical for virgin CDs and ones with marked rims.
Aside from all this, the claim is ridiculous. In fact, it is impossible.
The proposed mechanism is ludicrous. David Ranada has [thoroughly and experimentally debunked] the supposed optical explanations for the green marker phenomenon. Concrete data from CD audio players shows at worst one uncorrected error in every million bits. This means less than 200 of the 190 million samples on a CD will be in error. In practice there is more typically less than one uncorrected error for every two discs played, and more than 99% of the time these single errors will be detected and replaced by linear interpolation, losing only very high-frequency information and only for .02 milliseconds of music.
Even if digital read errors were several thousand times more frequent than they really are, so that undetected errors began to appear, they could not create the types of sound degradation supposedly fixed by marker treatments. Several reviewers for Stereophile magazine agreed that a good CD marker "increases soundstage definition, improves the solidity of bass reproduction, and usefully lowers the level of treble grain so typical of CD sound." Others refer to improved "clarity" and "imaging." All of these imply degradation caused by low-amplitude noise or by some sort of filtering process. This would require very small errors in almost every sample.
In contrast, bit errors at the read-mechanism level are just as likely to be high-amplitude as low, since they would affect high bits just as often as low bits. Thus infrequent errors from the digital train of a CD player would be audible as high-frequency clicks lasting less than a millisecond each. If the bit error frequency were gradually increased, the clicks would grow closer together, and then merge into an overlay of high-amplitude digital white noise like the sound of a data track played as audio. In no way could random errors in the digital audio stream account for the subtle, highly systematic degredation of the above claims.
(The errors would also be audible as clicks or fuzz in the zero-signal tracks on discs made for equipment tests like [this one]. They are not, of course.)