There's a long-standing belief, popularized by our dear Oswald Szemerenyi, that long vowels in nominative forms of Proto-Indo-European (PIE) nouns are caused by the loss of the nominative ending *-s after certain consonants (*m, *n, *r, *i). Yet there's just a few snags: *népōts 'nephew, niece', for one. It's accusative case form (used for the object of an action) is *népotm̥ and it's genitive form is *neptós 'of the nephew'. It seems that the long vowel is exclusively in the nominative case, that the nominative ending has not disappeared, and that the root is not monosyllabic! However, Szemerenyi explains this away by claiming that the assimilation of word-final *-ts in the nominative to *-s caused compensatory lengthening of the preceding vowel. And yet there's also the example of *dyēus 'sky' (confirmed by Sanskrit dyaus which, if the stem vowel were short, would presumably have become *dyos) and even though one might argue that the reconstructed accusative form *dyēm (> Sanskrit dyām) could likely derive from earlier *dyēum just as well as *dyeum, the genitive form *diwós shows that the underlying vowel can only be short *e by all the normal rules of PIE ablaut (i.e. long *ē alternates with short *e while short *e alternates with zerograde). If the underlying vowel of the stem were long *ē, then we would have expected genitive **deiwós instead (cf. Sanskrit divás) and yet the nominative form shows a long vowel anyway. This upsets the applecart nicely, don't you think?
This is why I choose to believe that the lengthened vowel of the nominative case is not caused by the loss of the nominative ending but rather by its prehistorical presence. I figure that the lengthening occurred right after Syncope when *-sa was "clipped" to *-s with additional compensatory lengthening. Normally, such monosyllabic suffixes of the form CV should retain their vowel according to the rules I've formulated so far, so this compensatory lengthening would be evidence of the nominative singular's unusual violation of the Suffix Resistance exception of Syncope. A kind of exception of an exception, if you will.
A while later, something totally unrelated happened in the language, the loss of *-s in certain paradigms. The last time I left the problem, I had deduced that word-final *-s was voiced to /-z/. This was a stubborn compromise I made after speaking with Jens Rasmussen on the Cybalist forum where he insisted that a phoneme *z existed in an earlier stage of PIE. I couldn't be moved to reconstruct an extra phoneme because it seemed to me to mitigate against peak theoretical efficiency (i.e. it was Occam's-Razor-unfriendly) but I had to concede that there was at least an added voiced allophone of *s at work.
This original voicing of word-final *-s appears to be indicated, for one thing, by the fact that the thematic vowel, which is known to alternate between *o and *e, becomes *o before nominative *-s instead of expected *e. Elsewhere *o regularly surfaces before a voiced consonant while it is *e that surfaces before an unvoiced one (e.g. *kʷer-o-més 'we would make' yet *kʷér-e-s 'you would make'). So the loss of the ending after certain coincidentally voiced resonants may also indicate the presence of an allophone /z/ that presumably merged back to /s/ in other environments such as after a thematic vowel. After voiced resonants, however, such a merger by devoicing would be difficult and so word-final /-mz/, for example, would instead become /-m/ (e.g. mid Late IE *dgāms /dga:mz/ > PIE *dʰǵʰōm 'earth').
 Szemerenyi, Introduction to Indo-European Linguistics (1999), Translated from Einfiihrung in die vergleichende Sprachwissenschaft, 4th ed, p.181 (see link).
 For a brief definition of the term compensatory lengthening and some interesting examples of it, see an abstract of Campos-Astorkiza, "A Typological analysis of compensatory consonant lengthening", Phonology and Phonetics in Iberia (June 2005) (or view the pdf here).