http://blog.lexa.ru/2013/07/14/pro_sraw.html
Ah, yes, thanks. Now it's coming back from my recollection. They are a 'lossy' YCrCb space encoded and decimated formats. So color accuracy suffers, but it of course depends on the subject matter if that is noticeable (especially without a better, downsampled reference).
It remains unclear how the subsampling of color is done, but most likely much less sophisticated than a good downsampling would achieve, perhaps they just simply average neighboring pixels, instead of taking a weighted average like with downsampling. Also the decimation method used for Luminosity (Y) is unclear.
Again, mRaw and sRaw can be perfectly usable for some applications, and may be the only way to store huge numbers of files on a limited (in camera) storage medium. But arguably better image quality can be had from proper downsampling of the full size Raw conversion. MRaw and sRaw are (only) about file size reduction, not highest quality.
Personally I would not use it as a means to reduce noise (if that even works adequately), because it's much more effective to do noise reduction at the full size and do proper down-sampling, and the noise vs detail trade-off benefits from having more pixels to base it on (even if the final output size is relatively small).
Cheers,
Bart