I only have a brief amount of time to delve into this and I promise that I will discuss this later but I've been recently contemplating the very thing I thought I would never contemplate. That thing is Jens Rasmussen's voiced sibilant *z in Pre-Proto-Indo-European (Pre-PIE) as was discussed on the Cybalist forum on Yahoogroups a couple of years ago. I would also like to thank Tropylium for sharing a comparison between the apparent word-final voicing in Pre-PIE that I mentioned in a recent post and Balto-Finnic gradation which also involves voicing based on syllable structure.
So, long story short, I'm devising an alternative account of my theory on Syncope and late Mid IE such that voicing of sibilants and stops arises in an unstressed sequence of word-final *-VCᵊ#. (The final superscript character is meant to represent a supershort schwa which would have existed just before Syncope deleted instances of them altogether.) In this scheme, voicing would only occur before a word-final supershort schwa and not before a word-final full schwa or other vowels in word-final position. Also, the preceding first vowel can be any vowel, including zerograded *i and *u which arose at precisely this time. The final syllable must be open as well and only one consonant can be involved in this sequence for it to work. I figure the reason for this restriction implies a relationship between the duration of the word-final vowel and voicing. I'll explain more later on how this would work with examples and how this changes a few things that I've held dear up to now. Maybe I'm still wrong, but I think this is worth exploring nonetheless.