Spirantization in Iranian, part 2
In addition to the universal voiceless stops *p *t *k, Proto-Iranian also had the voiceless affricates *c *č. Stops and affricates often develop in similar ways, for natural reasons (indeed, many theories of featural phonology posit that affricates in general are just a subtype type of stops). So I figure that the Proto-Iranian lenition of stops to spirants could have affected the affricates as well.
The “second” palatal *č did not have many chances to end up spirantized, since this had only evolved to start with from pre-Indo-Iranian *k (< PIE *k, *kʷ) before a front vowel. However, there was at least also the semivowel equivalent *kj > *čj — and this indeed predictably develops into *šj.
This leaves the PII “first” palatal *ć (from PIE palatovelar *ḱ), initially de-palatalized in Iranian to an alveolar affricate *c. Tracing possible conditional spirantization of this is challenging though, because soon after, it spirantizes unconditionally. This is linked to another notable sound change in Iranian, where first *s before vowels and most sonorants becomes *h (also, *sw becomes *xw, possibly but not necessarily through *hw). To fill in this partial gap, most Iranian languages shift *c to a new plain *s. Similarly also PIE *ǵ > PII *dź > *z. Compare e.g. Kurdish sed ‘100’ (<< *ḱ-), zer ‘yellow’ (<< *ǵʰ-).
An exception though is Old Persian. Here the development differs: *ć > *c > /θ/, *dź > *dz > *ð > /d/. One example still seen in Modern Persian is دست /dæst/ ‘hand’, from PIE *ǵʰes-. This is not as unusual as it may sound like: it can be compared also with the similar fronting of the PIE palatovelars all the way to th /θ/, dh /ð/ in Albanian, the fronting of Vulgar Latin *č, *dz to /θ/ in European Spanish, or the fronting of Proto-Uralic *ćć to *θθ in Old Finnish.
Now, back over at the Iranian system of consonant clusters… A few cases develop just as could be predicted, e.g. PII *ćr into *sr in “mainline Iranian”, versus *θr > ‹ç› in Old Persian. Some others clusters had been spirantized even earlier: *ćt ends up as *št already within common Indo-iranian (ditto *dźd > *žd), and the palatal quality of the consonant is preserved thanks to this. But at least in one case something funny happens. This is *ćw, which in mainline Iranian turns into *sp (e.g. Avestan aspa ‘horse’, from PIE *h₁eḱwos). Old Persian, by contrast, has in native vocabulary just plain /s/. Note that this last case cannot be through simple loss of *w! We would otherwise expect **θ.
Symmetrically with respect to voicing also: *ǵw > *dźw > *dzw > Mainline Iranian *zb, Old Persian /z/.
(OP has a few cases of /sp zb/ too, and Middle Persian even more, and thus e.g. Modern Persian اسب /æsb/ ‘horse’ — but these are normally attributed to loanwords from various mainline Iranian languages, such as Median, Parthian, Bactrian, Khwarezmian…)
So, suppose now that pre-Iranian *cw was too subject to spirantization in Iranian. If *s > *h had already happened, we would naturally get a new *sw, whose *s would then naturally survive in Old Persian (just as cases of *s survive in clusters such as *sp *st *sk).
On closer thought, it would seem like a better idea to assume apical [s̺] rather than laminal [s̻]. The former is the likely original value of *s in PIE and in most, perhaps all, of its early descendants.
It is regardless relevant to consider the apical/laminal division. The laminal sibilant would make a feasible intermediate between *c — perhaps laminal [t͡s̻], as is typical for affricates resulting from palatalization processes — and both of later Old Persian /θ/, mainline Iranian /s/. This would mean that while OP had taken a “Spanish” route in resolving the fairly rare /s̻ s̺/ contrast by fronting the laminal sibilant into an interdental, mainline Iranian had taken a “French” route by simply merging the two.
(Yes, by the way, early Middle French briefly distinguished an already deaffricated laminal c ç z /s̻/ from apical s ss /s̺/ before merging them. This stage can be seen in some loanwords into English, such as pousser → push, where /s̺/ was deemed more similar to /ʃ/ than /s/; or into Friulian, such as Louis → Luigi /luiʒi/, where the voiced equivalent /z̺/ was deemed more similar to /ʒ/ than /z/. Italian, in turn, then adopts the Friulian as /luidʒi/ within an affricate…)
So, the correspondence OP *s ~ other Ir. *sp clearly requires the apical sibilant. And if we think about things for a while, this development actually makes sense if we assume conditional spirantization to *sw before the general spirantization of *c. It is not realistic for conditional lenition of a single consonant cluster *cw to create a new, extremely marginal phoneme /s̻/! This would have simply merged with the single pre-existing sibilant /s/ [s̺]. To actually get an /s̻/ around, even fleetingly, would require something with serious contrastive or semi-contrastive load behind it — such as the unconditional spirantization of *c in all positions.
A stage of Old Persian where *c was generally reflected as *s̻ seems to clarify a few other correspondences too. I mentioned last time that earlier *tn *tj yield /šn šj/ in OP, and these likely first developed to *θn *θj. But how have these clusters, going dental to postalveolar, bypassed the stage /sn sj/, with the alveolar sibilant? In the latter we could assume palatalization *sj > /šj/, but inherited Indo-European *sn remains as-in in Old Persian.
My suggestion would be that while normally the laminal sibilant *s̻ merged in OP into /θ/, in these clusters the opposite happened. Interdentals are normally laminal to begin with, so *θ > *s̻ is perfectly reasonable. After these two changes, the laminal sibilant would be left marginal again. Instead of drifting forward once more, it could take the opposite route and become a postalveolar sibilant [ʃ] = /š/. Within much of Iranian *š seems to have clearly been a part of the general palatal series of consonants (together with *č *dž), and this would again mean that a laminal articulation is likely.
…Note, however, that this is in contrast with the Indo-Aryan branch, where *š becomes the retroflex [ʂ] = /ṣ/ (retroflexes are by definition apical). Also the Slavic development *š > *x likely requires a retroflex or at least some kind of a retracted intermediate. Lastly, since *š in the satem-type Indo-European languages first develops by the RUKI retraction of *s, it would likely also have started off apical to begin with.
