Under the heading 4.3.4. The Reflexive and the Reciprocal, the cited author named Winfred Lehmann who was coincidentally the former director of the University of Texas' Linguistics Research Center until his passing in 2007 writes: "As indicated above, § 4.1.5, the categories for reflexivity and reciprocity are commonly expressed with verbal affixes in OV and also in VSO languages. Only in SVO languages can we expect to find pronominal forms for use in expressing these categories." I've bolded the part of his statement that irks me a bit. The author subsequently explains how the mediopassive category of Proto-Indo-European (PIE) verbs is used and then finally more crazy ideas are claimed: "In PIE, as in many OV languages, there was no reflexive or reciprocal pronoun. In the dialects an adjective *sew(e)-, meaning ‘own’, was taken over as a reflexive adjective; it was also adapted as the basis for the reflexive pronoun." Eeeek! Here are the problematic assumptions I see:
1) Bizarre reconstructions like 1ps subjunctive **-oxoi and 1ps optative **-yeʔxo are first assumed.
2) A mediopassive affix **-o- is extrapolated from #1 for PIE (and Pre-IE).
3) It's claimed that only SVO languages are expected to have reflexive/reciprocal pronouns.
4) Since PIE was not SVO and based on #2 and #3, *swe cannot be a reflexive pronoun.
5) Since it's not a pronoun, it cannot have originally meant 'self'.
6) It must mean 'dear' based on the Greek example of philos.
Holy non sequitur, Batman! First of all, the 1ps subjunctive is typically understood to simply be *(-o)-oh₂ (although Jasanoff convincingly argues for a purely "athematic" *-oh₂ in the earliest stage of PIE, contrasting with present indicative *-mi) and the 1ps optative is normally *-yeh₁m. So Lehmann first forces us to swallow his unique brand of PIE. Since Anatolian, Italic and Celtic share a mediopassive in *-r due to their geographic proximity to each other during their development in the PIE speech area, it's immediately clear why a language would expand the usage of the ending *-i seen in the active to the mediopassive by way of paradigmatic levelling. It's not clear however how or why a language would adopt a suffix *-r ex nihilo specifically for the mediopassive and replace a former *-i. This is why many IEists bow to common sense and recognize that the r-mediopassive is the true archaicism here, hence 1ps mediopassive *-h₂ór. There simply is no affix **-o- seperate from the following *r and it needs to be explained why there is no sign of this *-ór element in the 1pp and 2pp counterparts. Already Lehmann's theory tormented by his hyperanalysis is falling apart.
As for assumption #3, the SOV languages that I'm aware of do indeed have reflexive morphemes other than affixes but some may argue pedantically about what qualifies as a "reflexive pronoun". Whatever. The point however is that the unmarked word order of the language (whether SOV, SVO or VSO) has no bearing on the presence or absence of a reflexive pronoun or its equivalent. In Japanese, there is zibun; in Burmese, there is ko; in Turkish, there is kendi. All of these languages put the verb at the end of a sentence and all of them avail themselves of morphemes other than verbal affixes to convey the reflexive. So again, I fail to understand Lehmann's train of reasoning.
Given this, we can't justify the wholesale dismissal of the PIE reflexive pronoun (or whatever one prefers to call this morpheme). Regardless, *swe is still reconstructable for PIE and it still demonstrates a reflexive sense throughout the entire family from Germanic to Indo-Iranian despite the coexistence of a mediopassive. It's proposterous at this point to claim that all these branches somehow developed the meaning of 'self' in tandem without this all deriving from an available reflexive word in PIE! Talkin' about denial.
What we might come to conclude is that Lehmann put the cart before the horse and where the wild passion of his meandering argument lied was in comparing the semantics of Greek philos to PIE *swe. Yet this, like all the other claims are empty assumptions that haven't been proven. They're merely impressive connections that ignore so many other better possibilities in the process. One need not have to think long and hard about how a reflexive word can used to good use even in a language with a mediopassive that often covers reflexive actions. Consider for example a verbless phrase like "One's own horse." Can someone please tell me how to avoid using *swe or any of its related forms in this context? Or what so many other contexts where reflexive pronouns can add value?
Then there is the matter of the etymology of *swesor- "sister" which has already been determined rather brilliantly to derive from *swe and *-sor-, the same feminine ending seen in the ancient stem *kʷetwor-sr-es, the feminine form of "four" preserved in Celtic languages. The word "sister" probably literally meant "one's own girl" and may reflect patriarchal attitudes in prevailing Indo-European cultures at the time concerning patrilineality, kinship, and societal roles.
In light of this, to say that "statements in the handbooks ascribing a reflexive pronoun to PIE must be revised" is hasty bunk. Don't dare touch my *swe, people!
 By "athematic", I mean this in terms of traditionally theorized PIE grammar. In Jasanoff's view however, thematic verbs (e.g. *bʰér-o-h₂ 'I carry') are underlyingly the original subjunctive forms while athematic verbs (e.g. *h₁éi-mi 'I go') are archaic presents. Over time, Jasanoff presumes that as subjunctives became presents in their own right, they developed "doubly marked" subjunctives (hence the original subjunctive *bʰer-e-t(i) was replaced by a hypercorrected *bʰer-e-e-t(i) -> *bʰer-ē-t(i)). Thus by comparison with the traditional "thematic subjunctive" *-o-oh₂, Jasanoff's new interpretation of *-o-h₂ is relatively "athematic" although ironically the *-o- here is still the distinctive thematic vowel of the subjunctive seen in the "athematic verbs" of the traditional theory. Sorry for the confusion but this is a confusing topic that's hard to explain without juggling between two very different accounts of PIE grammar, traditional and progressive.
 This theory was published by Benveniste in 1969 (link).
(Mar 05 2008) I decided to add footnote #1 because my use of "athematic" is ambiguous here. Also updated **kʷetwe-sor- to proper *kʷetwor-sr-es in light of OInd. cátasraḥ. I guess my memory is failing me. Good thing I crack the whip on myself and verify details like this. Oy veh! (To add though, this "feminine" numeral form is certainly post-PIE because feminine gender is not reconstructable for the earliest stage of PIE and was no doubt built on a noun *sor- "woman" existing at the time.)
(Mar 05 2008... later on that evening) Damn! I've been punked, yo! By IEists! Szemerenyi reconstructed *kʷete-sor- (1977) in Studies in the Kinship Terminology of the Indo-European Languages (see link). No wonder I'm confused. Oh well. This is all just fun trivia because this post-IE numeral form really has nothing to do with PIE itself anymore. But now I know I'm not forgetful, arrrgh! :)