If we take for granted that *-i in Common PIE was once only an aspectual marker restricted to the realis mood and signalling an ongoing action (regardless of whether occurring in the past, present or future), then we must treat Hittite's contrast in the hi-class between past and present-future, which mirrors the same development towards tense in the mi-class, as an innovation whose source lies in PIE's "punctive" conjugation using only the tenseless *h₂e-set of pronominal endings. However, if true, why does the 3p of what we might be tempted to call a "perfect preterite" have a completely different form from the other two persons? That is, while the present tense in Hittite reflects a canonical 3ps form *CóC-e-i which can be trivially explained away as the mere addition of continuous-turned-presentive marker *-i in post-PIE times, the 3ps preterite form showing instead *CēC-s-t (rather than expected *CóC-e) is a real mindbender, particularly so with its added change in root vowel too! Hunh?!
However, referring back to my previous-stated model of IE conjugation, we can tease out a really simple reason for all of this odd innovation. As I stated before, we have the following basic forms of various aspects, all lacking any explicit reference to tense and based fundamentally on contrasts involving continuous/non-continuous (ie. whether the action or state has a defined start-/end-point or not) and punctive/non-punctive (ie. whether an action or state is instantaneous or enduring):
bʰēr-t '(S)he carries/(S)he carried (habitually)' (non-continuous)
bʰēr-s-t '(S)he carried (once)' (experiential)
bʰḗr-ti '(S)he is carrying/(S)he was carrying' (continuous)
wóid-e '(S)he knows/(S)he knew' (punctive)
We can now readily see from the above list that Hittite's 3ps preterite is identical to the experiential form (> sigmatic aorist), complete with the long *ē vocalism that tags non-punctive verbs. The interesting thing about the experiential aspect as it's used in Mandarin is that, being that it expresses an experience someone has had, it can only rationally refer to a point in time in a past relative to the timeframe conveyed by the speaker. It's semantics are squarely [-present] by definition, despite not being intended explicitly as a past tense marker per se. Similarly then, a form like *CēC-s-t with added *-s- simply signals that a naturally ongoing action (such as swimming or breathing) is now referring specifically to a singular point in time ([+momentaneous]), an experience in a person's life. We can then take note of an interesting aspectual contrast between *bʰḗr-m̥ 'I carry/carried' with no specific event being conveyed (potentially habitual), and the semelfactivizing quality of the sigmatic form *bʰḗr-s-m̥ 'I have carried (once)', acting essentially like a perfective for inherently durative verbs. Notice also the root vowel with lengthened *ē. So the preterite of the hi-class is nothing more than a matter of a former sigmatic experiential, already with momentaneous meaning, replacing the expected form *CóC-e.
Now the question left is why the bizarre replacement of o-grade 3ps with a lengthened e-grade form with an entirely different ending, seemingly on the side of increasing system irregularity? This, I believe, is the easiest question to answer involving both the semantics of the replacing form and the comparative failings of a system that would have resulted without this change. We need only compare the resulting hi-class presentive *CóC-e-i with a would-be preterite **CóC-e to understand that the change certainly helped to better phonetically distinguish between two almost homophonous forms. And too, it helped that this "s-aorist" already had the quality of momentaneousness built right in (ie. conveying a single point in time), blending well with a punctive conjugation that naturally did the same. Later, as the system settled into a new tensual contrast, any aspectual distinctions between the mi-class and hi-class dissolved since the only thing that mattered now, grammatically speaking, was past and present-future (ie. when an action occurred), not the aspect (ie. how an action occurred).
(13 Aug 2009) Just noticed something I might want to rearrange with better clarity: "We can then take note of an interesting aspectual contrast between *bʰḗr-mi 'I am/was carrying' with no specific event being conveyed (potentially habitual), and the semelfactivizing quality of the sigmatic form *bʰḗr-s-m̥ 'I have carried (once)' [...]" I should more aptly compare the non-continuous form *bʰḗr-m̥ 'I carry/carried' to the sigmatic "aorist" since they both end up being employed for past tense in later PIE dialects unlike the continuous (> presentive). There. Changed. To appreciate the subtlety in a tenseless model like this, *bʰēr-t without *-i must have originally expressed the act of carrying without any particular event defined (even one that may not exist, such as in the negative, *ne bʰēr-t '(s)he doesn't/didn't carry', or in the subjunctive with the added modal affix, *bʰér-e-t), while *bʰḗr-ti by contrast established at the very least that the action extended into the present (or rather, the "present" relative to the surrounding timeframe established by context). We should be able to see how easy it is for this to develop into tense, even for dialects to develop their own parallel tense systems independently from this single model.
(14 Aug 2009) Added a second reference to footnote #1. Replaced Hittite example of dāis with nāis 'he led'.
 Koch/Bowern/Evans/Miceli, Morphology and Language History: In Honour of Harold Koch (2008), in Amsterdam studies in the theory and history of linguistic science, v. 298, p.157 (see link): "The unexpected agreement between Tocharian and Anatolian in this detail lies at the heart of the 'h2e-conjugation' theory of the s-aorist presented in Jasanoff (2003: 174-214). According to this theory, the PIE s-aorist was originally a specially inflected type of root aorist in which the 3 sg. active form, for reasons now lost within the prehistory of PIE, was built from a suppletive sigmatic stem with 'Narten' (*ē : *e) ablaut."; Polome/Winter, Trends in linguistics 58 - Reconstructing languages and cultures (1992), p.143 (see link), concerning the 3ps preterite forms shared between Tocharian and Anatolian.