My cursed snag to my subjective-objective theory remains that subjectives are expected to yield atelic verbs (ie. verbs without completed goal) or imperfectives (ie. continuous actions or states). Even though the middle clearly should be placed under a subjective category in my hypothetical remodeling, and even though the subjective on which the middle is built acts in many respects like a former subjective (for example, its curious habit of accomodating verbs of state), subjective can't turn into a perfective in any direct way according to known structural linguistics! We oddly expect the total reverse: subjectives yielding imperfectives and objectives yielding perfectives. This is why I had a migraine a couple of days ago, frenetically diagramming to make this irritating paradox finally unfold into clarity. I think I've come to an interesting idea that takes a page from the Ancient Egyptian verbal system: eventive versus non-eventive.
An eventive verb, as the name implies, refers to a specific event in time while the non-eventive by contrast focuses more on the action or state in a more generalized context. It's the difference between "I ate the baby (last night)" (eventive) and "I eat babies (in general)" (non-eventive). By drawing a two-dimensional grid between eventive & non-eventive on the one hand and subjective & objective on the other, we end up with four main categories in which to place the earliest verbs of Common Proto-IE. This is perhaps not a good synchonic model for PIE but it seems to be explanatory as a seed for its eventual evolution in many of the Core IE dialects. Observe:
|Objective non-eventive||*h₁és-m̥ 'I am'|
*déh₃-m̥ 'I share'
*bʰḗr-m̥ 'I carry (often)'
|Objective eventive||*bʰḗr-mi 'I am carrying'|
*bʰḗr-s-m̥ 'I have carried (once)'
|Subjective non-eventive||*wóid-h₂e 'I know'|
*ḱónk-e 'it is hung'
*stestóh₂-e 'it is stood'
*gʰónh₁-h₂e 'I am born'
|Subjective eventive||*h₁és-h₂or 'I sit'|
*gʰu-h₂ór 'I spill'
This category is the origin of root aorists and imperfective past in Core IE while becoming the mi-class preterite in Anatolian. This is because the pan-PIE choice to make the continuous aspect a present tense marker made non-continuous verbs automatically a past by contrast. This change also hints at why past tense came to be less marked than the present, curiously opposing normal language tendencies (ie. the shift occurred rapidly as PIE broke up). Telic verbs (ie. Narten presents with *ē-vocalism) were free to be marked in the continuous, suggesting an impending goal, or to remain unmarked to convey realized ones. The former leads to a durative-turned-present and the latter leads to a momentaneous-turned-past.
This category was largely characterized by telic verbs marked in the continuous aspect (later Narten presents). However, the antecedent of sigmatic aorists (ie. those verbs marked in *-s- with lengthened root vowel) which originally expressed a past experience were by definition eventive as well. An experiential form, parallel to Mandarin guo (过), can easily yield explicit past tense "sigmatic aorists" in Core IE dialects, while forming special 3ps sigmatic past forms for an originally tenseless hi-class as evidenced by Anatolian and Tocharian.
This category denoted states where the subject was affected (passives), but also often verbs of emotion and of thought which are clearly subjective, as well as what would later be termed reduplicated perfects. The development of reduplicated perfects with built in punctual meaning directly out of a "stative" requires the brunt of explanation. My hunch here is that reduplication was restricted to verbs like 'stand' signifying states resulting from prior action. An expression like 'it is stood' qualifies as such while 'I have gone' is less focussed on state than on the completed action itself. This latter comparison explains how forms specialized in suggesting resultant states from actions (PIE *stestóh₂e 'it is stood up' → 'it has come to be stood up') can gradually expand to take on resultant actions as well (post-IE *bʰebʰórh₂e 'I have come to be carrying' → 'I have carried') and thus become a "perfect" with punctual aspect now emphasizing active result over mere resulted state. The connection then between subjective and perfect is indirect and subtle.
It should also be noted that only part of this single category yields perfects while the rest of the subjective umbrella is the mother of middles and inherently non-punctual states. So the subjective simply doesn't become perfective wholesale and there's no need for me to worry about the crosslinguistic tendency I mentioned above which refers more to developments as a whole, not in part. Phew!
Also, forms like *wóidh₂e 'I know' which never ever show reduplication in the later perfect hint at their original meaning and usage: 'I know' (stative) → 'I have come to know' (inchoative) → 'I have known/seen' (perfective past). Since 'to know' is not an action and since reduplication expresses a resultant state from an *action* as outlined above, naturally there can be no reduplicated forms possible for these stative verbs.
This category is composed of middles and are marked as such with special endings in *-r. They are actions accomplished through a medium ("mediopassive"), through one's own effort ("sitting"), or involuntarily like "spilling" or "sneezing". This category appears to be the least molested by change in the various branches, so I think I'll stop typing now.
There, now we'll see if this post stands the test of time...
 Gülzow, The Acquisition of Intensifiers Emphatic Reflexives in English and German Child Language (2006), Studies on language acquisition, v.22, p.45, table 6 (see link).
 Perhaps this verb is better placed in the category of subjective non-eventive instead. Note Hewson/Bubeník, From Case to Adposition: The Development of Configurational Syntax in Indo-European Languages (2006), p.100 (see link): "In this context we should remind ourselves of the archaic transactional meaning of *dō- (< *deh₃-) as shown by Hittite dahhi 'I take' (< *deh₃-h₂e) vs. general IE 'I give'." Jasanoff shows a slightly different interpretation in The role of o-grade in Hittite and Tocharian verb inflection (1992), published in Reconstructing Languages and Cultures, p.140 (see link): "Hittite has a 3 sg. dāi, which can be referred to an underlying *doh₃-e and taken as the continuant of a pre-Proto-Indo-European 'middle' with the meaning 'gives/gave to oneself'."
(12 Sep 2009)
Just realized that a better translation of what must have been a bidirectional sense underlying *deh₃- is not 'to trade' but 'to share'. Thus, a semantic shift from 'to share (my own property)' → 'to give' (as in Core IE dialects) and 'to share (another's property)' → 'to take' is clearer. It also works well with what grammatically hints at a non-eventive and atelic verb, non-eventive because it avoids the continuous marker and atelic because it shows no Narten-grade which would explicitly mark it as telic.
Also corrected the forms *h₁ḗs-h₂or and *gʰeu-h₂ór to *h₁és-h₂or and *gʰu-h₂ór respectively, as per the middle forms suggested in Sihler, New comparative grammar of Greek and Latin (1995), p.133 (see link) and in Indo-European perspectives - Studies in honour of Anna Morpurgo (2004), p.506 (see link).
(05 Sep 2009)
Added footnote #2 concerning the changing semantics of *deh₃-.
(09 Sep 2009)
Added footnote #1 concerning theories on the typical evolution of transitive verbs.