Having spent too much time pondering the subject, I have determined that for some reason I want my next big conlang project to be not a language but essentially a whole family of languages. It's been done, but that's no reason to stop myself, right? Kalis is not going away and it still needs some polishing, but this is calling me. And furthermore, I want it to be vaugely Indo-European.
So first, I need the ancestor language. Time to pick phonemes!
I'm not going to just rip off the PIE system. I don't think. For one thing I can't even pronounce the PIE system, and I take it entirely on faith that there really is a difference between, e.g., aspirated and unaspirated P--I can hear the difference between, say, alveolar and retroflex consonants (otherwise known as "Why native Hindi speakers pronounce T and D funny according to English speakers"), but aspiration is a completely closed book to me. Also, if I did use the PIE system the odds are good that any root I invented would already be a root in PIE, though with an utterly different meaning, and this would likely cause me some cognitive dissonance. So I must think on this.
One of the reasons PIE's daughter languages diverged so much, I imagine, is that the original system had a whole bunch of really fiddly distinctions--we're talking "three different, phonemically significant versions of K and six of G" here. Nine phonemes, one place and type of articulation (velar stop, to be precise); is it any wonder the daughters spread out through the phonemic space as much as they could?
I have to think about it.