In Quenya the dental voiced spirant [ð] and the voiced sibilant [z] eventually merged (to [r] in the Ñoldorin dialect and to [z] in the Vanyarin dialect). Before this merger was accomplished, however, these sounds dissimilated away from any nearby r whose pronunciation they had begun to resemble: to l in the case of ð and to s in the case of z:

• r-ð > r-l
• r-z > r-s (and z-r > s-r).

It seems the dissimilation of z > s only occurred in the Ñoldorin dialect, and this must have occurred after þ > s reintroduced this sound medially into the language. The dissimilation of ð > l occurred earlier and in both Vanyarin and Ñoldorin speech. It was probably similar to (and perhaps contemporary with) the change whereby initial [d] became [l]. The Outline of Phonetic Development (OP1) from the 1940s only mentioned this r-ð > r-l dissimilation:

Medially l [from primitive *d] also occurs, but this is apparently by dissimilation since it is seen only where r already occurs in the word: thus *r-d regularly becomes r-l (OP1: PE19/32).

This sound change has essentially the same description in the Outline of Phonology (OP2) from the 1950s:

Medially l also occurs as product of older d. This is most frequently due to dissimilation, in such sequences as rad- > ral- (not rar-) (OP2: PE19/70).

The timing of this sound change is not clear. It may have occurred at the same time as the initial d- > l- (as noted above), or it may have been later. It was definitely before intervocalic [ð] became [r]. In the Eldamo data model I put this sound change in the Parmaquesta [PQ] period mainly so it can be discussed in conjunction with the other dissimilation r-z > r-s, but it quite possible it happened in Ancient Quenya [AQ].

Apparently Quenya had a general dislike for sequences of r, as noted in OP2:

Substitution of l for r was in such cases more frequent than the reverse, since Quenya in general preferred l, and did not object to sequences of l, but avoided the sequence r-r: there were no bases of that form, and even such a base as CE √RAD became ral-, not rar̃- > rar (OP2: PE19/99).

The avoidance of r-r was not universal, though:

But r-r did occur in some derivative words, especially those in which r had some phonetic or onomatopoetic significance: as raurō “lion”, beside older CE rāw, archaic Q rá. Cf. Rauros(se) “roaring-rain” — name of a large, loud waterfall (OP2: PE19/99, a footnote to the above).

Another interesting example is the word Q. lala-, which seems to indicate ð could also assimilate to a preceding l, giving another case where medial d > l rather than r:

• ✶g-lada- > Q. lala- “to laugh” (PM/359).

However, there is also ᴹQ. lára “flat” < ᴹ√LAD (Ety/LAD, DAL), so this assimilation to preceding l was either a late idea or was not universal.

The dissimilation of r-z > r-s was only mentioned in OP2, and was a later addition in red ink (PE19/74 note #31):

When r preceded or followed z (even with an intervening vowel) the rhotacism did not take place in Ñ[oldorin] or Tel[erin]. In such cases z was again unvoiced to s by the Noldor in Beleriand. So raze (rase) “sticks out” (not rare). In this case medial d/s are distinguished — since d here became l (OP2: PE19/73-74).

As indicated by this note, the second dissimilation occurred in late TQ (post-Exile). Aside from the example above, this sound change can be seen in:

• Kh. Khazād [> *Kazar] > Q. Casar “Dwarf” (PE17/45; WJ/388).

This adaptation must have been post-Exile, since the Noldor did not encounter the Dwarves until they reached Beleriand. One interesting thing about this word is that the sound change ð > r seems to take priority over z > r, indicating the phonetic development of final ð > r was older. Indeed this sound change occurred even in the Vanyarin dialect in the final position (PE19/104). The phonetic development of Casar and the above note from OP2 also indicate that the sound change z > r may not have been fully realized in Ñoldorin Tarquesta until after the Exile.

Conceptual Development: Words with r-r are fairly common in Early Qenya of the 1910s, so it is unlikely these dissimilations applied to the earliest conceptual stage of the language. In fact ᴱQ. rēro “cousin” was derived from the root ᴱ√REÐE [reðe] (QL/79). There is also a counter-example of d > ð > l after r appearing in The Etymologies of the 1930s: ᴹ√RED > ᴹQ. rer- “to sow”. This is consistent with it being a sporadic rather than universal change (PE19/99).