There seems to be profuse confusion over the quality of medial -r- in Early Middle Chinese. Most people agree that it is difficult to put a phonetic value on -r- in Old or Early Middle Chinese - it might be a flap/tap, trill, alveolar approximant, velar fricative, etc. It's fine to be agnostic. However, some people are more ambitious and would like to put a finger on the precise realization of this phone.
Particularly amusing to me are those hypotheses which have -r- being realized phonetically as a voiced velar fricative or the like. This is fine and all until you wake up to the two following facts: One, voiced velar fricatives - and their relatives, voiced velar/uvular trills/fricatives/approximants - generally don't produce coloring of neighboring vowels. Two, if any coloring, the coloring would be a velarization or lowering of the second formant of the colored vowel.
Of course, Middle Chinese second division vowels are all low front vowels. Go figure.
Actually, I find the agnostic view bothersome as well. I think it's quite obvious from the two effects of medial r - inducement of retroflexion of coronals and fronting of vowels - that it has to be a coronal approximant of some sort. It has to be a coronal otherwise it wouldn't be able to affect coronal consonants, and it has to be an approximant, as trills and the like generally don't color neighboring vowels.
Then again, one could always just cast their eyes towards neighboring languages. Hmongic languages have a ʒ or ʐ reflex of proto-Hmong-Mien *r (in Chinese loans as well - witness Old Chinese *mə-roŋ > White Hmong zag /ʐa2/), and proto-Tai probably had a realization of *r as /ʒ/ due to its various reflexes as /h/ and such.
Furthermore, the Hmongic languages probably preserve pronunciation patterns from way back when, since they migrated south before the Tang dynasty, and stayed relatively safe from contact with the Chinese until much later; proto-Tai was a language very much in contact with Old Chinese.
Add to the contact situation/migration patterns the fact that the phonetic realization of rhotics spreads areally very easily in close contact situations.
Granted, most of the points are somewhat speculative and don't hold much water by themselves. But they all point to more or less the same conclusion: Old Chinese *r was probably pronounced like /ʐ/. What else would it be, anyway?
