"Most commonly, the 'specific structural purpose' of regular metathesis lies in converting phonologically or perceptually 'marked' structures into more acceptable ones."This, of course, is exactly what I've been proposing in relation to the Syncope rule in my theory of the transition between late Mid IE and early Late IE. The mass deletion of supershort schwas throughout the language would have put strong pressures on syllable structure, producing complex clusters where there originally were none. From the looks of it, early Indo-European initially placed strict limits on what clusters were legal such that word-initial CCC- was virtually banned aside from clusters preceded by *s- (e.g. *spl-, *str-, etc.) and, yes, this means that I think that *ph₂ter- "father" and similar words are datable only to the post-Syncope Late IE period. That metathesis happened in pre-IE is, as far as I'm concerned, a fact proven by the infix *-n- which was surely at one time a suffix *-an- in Mid IE (n.b. suffixes are most typical of SOV languages) before it started doing a doe-see-doe with the last consonant of the verb stem, coincidentally sometime during Syncope (e.g. *leikʷ- + *-n- = *li-n-kʷ-).
Now that we got that straight, the same book continues on to state:
"Regular metathesis may also serve to eliminate clusters which do not conform to the preferred ordering of segments within syllables in terms of the sonority hierarchy (cf. 2.1.12 above). For instance, in the prehistory of Modern Persian, apocope introduced word-final clusters of obstruent or nasal + [r]. The increasing sonority of these clusters violated the tendency toward decreasing sonority of the coda. Metathesis eliminated the 'marked' structures and replaced them with 'unmarked' ones."Eureka! Thank you, Old Persian, I love you. My point is that, while some of my commenters here have suggested that my hypothetical post-Syncope verb stem **rten- in my previous post (see Phonotactic processes during Syncope) could have just as easily gone through a Sino-Tibetan-like deletion of offending *r-, this is not the only evolutionary possibility. Metathesis of the liquid can indeed happen to produce something like **tren- as I suggested and for the very reasons I suggested. In fact, one might even say that deletion is less preferable than metathesis because the former process destroys the inherited consonantism and can reduce intelligibility.
There's more. As my mind was voyaging across the internet the other day, I came across a rather kooky book that sadly was trying to revive the credibility of a linguistic connection between the Indo-European and Semitic language families (a rather silly pet theory in this day and age). It was Saul Levin's Semitic and Indo-European where it hooked me with these lines:
"Another root, apparently triconsonantal in Semitic but biconsonantal in IE, involves a metathesis of the first and second consonants:" (bolded text my own)Say what? Did someone say "metathesis"?! As much as the tired Indo-Semitic theory presented in this book is an immediate write-off, I still think that some of these ideas may be of use (little use, but a use nonetheless) in hinting at what would more plausibly be evidence of borrowings and contact between early stages of Proto-Indo-European and Proto-Semitic rather than of outright linguistic relationship. So the equation that Levin provided here (i.e. Semitic *rdm and PIE *drem- both meaning "to sleep") has got me thinking in light of what I'm now aware of about pre-IE metathesis. Is it possible that we're dealing with a Semitic borrowing adopted into Mid IE as *rad̰éma-, thereafter shortened to *rᵊd̰émᵊ- by Reduction and transformed to *d̰ᵊrémᵊ- by phonotactically-motivated Metathesis before finally becoming early Late IE *d̰rem- by way of Syncope?
Of course, one example doesn't prove much at all, but perhaps this is a start of something. Cross fingers and pray for my sanity...
 Hock, Principles of Historical Linguistics (1991), p.115 (see link).
 Levin, Semitic and Indo-European (1995). Amsterdam studies in the theory and history of linguistic science, v. 129, 226. pp.21-22 (see link).