In Quenya, short final ă, ĕ, ŏ were lost in long compounds. Tolkien discussed this in the Outline of Phonetic Development (OP1) from the 1940s:

At end of long words, especially compounds, ă, ĕ, ŏ, reductions of ē, ā, ō (as described above), were also lost. Thus kwènedḗ > kwendḗ but móri-kwènedḗ > móri-kwènǝdĕ > móri-kwèndĕ > móri-kwèn(d), Q. Moriquen.

These compounds were often reformed from the old simplex; and sometimes the reverse took place: a form proper to a compound was extracted and used separately. Thus beside Moriquen also Moriquende. An interesting case is Q Vălĭnor beside Vălĭnṓre. This is derived from Bắlī + ndṓrē “land” > Bálīndṑrē, Bálĭndṑrĕ, Balindor > Q Valinor reformed after the simplex nōrē but preserving the ĭ due to ancient nd. The compound is old as shown by this formation (OP1: PE19/59).

These final vowel losses resemble the vowel losses whereby short final [e], [a], [o] vanished in Primitive Elvish, but this second wave of vowel losses was probably later, as seen by the fact that they occurred after certain Quenya-only sound changes such as how medial [s] often became [z]:

• ✶Eldă-kāzā > Eldă-kār(ă) > Q. Eldacar “Elf-helm” (PE17/114).

This example seems to imply the final vowel losses also took place after [z] became [r], but other examples show that [z] > [r] only occurred after final consonant clusters reduced:

• ✶tindōmi-sel(dĕ) > tindómizel > Q. Tindómerel “Twilight-daughter” (PE19/73).

I think it is likelier that these Quenya-specific short-vowel losses occurred after intervocalic s > z but before z > r (a change occurring very late and in the Ñoldorin Quenya dialect only). It’s possible, however, that these vowel losses in long compounds remained an active sound change for a very long time. These long/short compound pairs like Moriquende/Moriquen and Valinóre/Valinor are quite common in Quenya. Given the Elvish tendency to (a) use compounds for person names and (b) dislike person names of more than four syllables, I think such reductions of the final elements of compounds continued to occur even into the Tarquesta [TQ] period, but probably began in the PQ period or possibly even earlier.

Note that in most (if not all) cases where these short vowel losses occurred in Quenya, they were originally long vowels, which means these short-vowel losses must have occurred after the long vowels shortened. However, as discussed in the entry on how long final vowels were shortened in Quenya, the long vowels at the end of compounds were shortened much earlier than in simplexes, and so it doesn’t give much of information on when the short vowels losses might have occurred.

Conceptual Development: It seems likely these reductions date back to the Tolkien’s conception of Early Qenya, but it is hard to tell the difference between primitive vowel losses and later vowel losses, so there may not have been two phases of losses in the 1910s and 20s:

To this period [Primitive Eldarin] belongs the origin of -ə and probably also the complete vanishing of ŏ, ă, ĕ, through ə in certain forms ...

Long vowels final were very common in Eldarin. Short were practically only u, i, ə. [In Cor-Eldarin] These longs now became short; u > o; i > e; and ə vanished (PE12/4).

It does seem, though, that Quenya gradually became more resistant in Tolkien’s mind to final vowels losses that did not result in “acceptable” final consonants. In the Outline of Phonology (OP2) from the 1950s Tolkien said:

This list of “permissible finals”: n, r, l, s, t and nt remained constant in Quenya speech-feeling. Weak vowels were most readily lost when their disappearance left simple n, r, l, s, t finally (OP2: PE19/104).

In Early Qenya, there was much more variation in stem forms versus uninflected forms, showing a greater tendency to lose final vowels even when the result was a monosyllable. Many of these forms reappear in later writings with a final vowel:

• ᴱQ. an (and-) “gift” (QL/31) vs. Q., ᴹQ. anna (PE17/90, Ety/ANA¹).
• ᴱQ. let (lept-) “finger” (QL/53) vs. ᴹQ. lepse (Ety/LEP), Q. leuta (PE19/84).
• ᴱQ. lin (lind-) “air, tune” (QL/70) vs. ᴹQ. linde (Ety/GLIN).
• ᴱQ. nan (nand-) “woodland” (QL/64) vs. Q. ᴹQ. nanda (PE17/80, Ety/NAD).
• ᴱQ. on(d) “stone” (QL/70) vs. Q., ᴹQ. ondo (Let/410, Ety/GOND).
• ᴱQ. Qen (Qend-) “Elf” (QL/92) vs. Q., ᴹQ. Quendë (WJ/361, Ety/KWEN(ED)).
• ᴱQ. tet (tekt-) “spot, mark” (QL/90) vs. Q., ᴹQ. tehta (PE22/149, Ety/TEK).

This may mean that Tolkien decided primitive forms were more likely to end with a long rather than short vowel, but this trend is so ubiquitous that it could imply a tendency to retain final vowels whose loss would produce “unpleasant clusters” in later iterations of the language, especially in short words. Many of the above still reduced in compounds up through the 1950s and 60s, however (-lin, -nan, -quen).