Tolkien described several phonetic developments for the Quenya combination sr in the history of the language. He described essentially the same phonetic developments in both the Outline of Phonetic Development (OP1) from the 1940s and the Outline of Phonology (OP2) from the 1950s:

While s was still voiceless sr > sř > ss (similar to developments after stops) ... Later forms have zar (OP1: PE19/49).

In that case there are three developments of sr: (a) oldest, while s remained voiceless, sr > sř > ss: this commonly survived, being sufficiently perspicuous etymologically, in PQ and TQ; (b) after the voicing of intervocalic s > z, sr > zr > rr (such forms were more frequent in Ñ[oldorin] dialect, since z itself had there become r); in late formations z-r > zar (OP2: PE19/102).

The “intermediate” sound change (sr > zr > rr) appears in the first draft of a table of medial developments in the 1930s (OP1: PE19/52 note #119), but this page was rejected and the phonetic development does not appear in the final table, although sr > ss does (OP1: PE19/51). It seems the series of three distinct sound changes for sr was restored in the 1950s.

The first of these sound changes (sr > sř > ss) probably belonged to the Ancient Quenya period. Despite the supposed “survival” of this ancient sound change, I can find no examples of it in the phonetic history of individual words, where the intermediate voicing change of sr > zr > rr is more prevalent:

• ✶cas-raya [> *cazrea] > Q. carrëa “treasure” (VT42/12).
• ✶mi-srawanwe [> *mizroanwe] > (plural) Q. Mirröanwi “Incarnates” (MR/350).
• ᴹ√SRŌ [> *ozro] > ᴹQ. orro “up in the air” (PE22/127).

Thus, the ancient sound change of sr > ss was either frequently reformed, or Tolkien abandoned this phonetic development when forming actual words.

Conceptual Development: The one clear example of sr development in Early Qenya also shows sr > rr:

• ᴱ✶res-rŭ > rerro > ᴱQ. rēro “cousin” (QL/79).

Neo-Quenya: For purposes of Neo-Quenya, I think it is best to ignore this ancient change and assume that sr > zr > rr was the normal development, as described in the entry on how [z] and [ð] assimilated to following [r], [l].