Based on various examples (mainly from Noldorin), primitive polysyllables ending in final voiceless stops (p, t, k) or s lost this final consonant in their phonetic development. Such primitive words were principally derived from KALAT-stems and fairly rare, especially those ending in [p] (PE21/71). Tolkien discussed such words in his notes on Common Eldarin Noun Structure in the late 40s:

Examples appearing in more than one language and so probably ancient are ... nelek “tooth”, philik “finch”, usuk “dusk, evening”, kjelep “silver”. Owing to loss of final consonants, and in some branches (such as Q.) of the vowel of the second syllable by syncope in inflected forms, this class of nouns was much reduced in later Eldarin tongues ... {ON >>} OT perhaps preserves the old formations best, as in uso (pl. usuki); nele (pl. neleki); phile (pl. philiki) beside phlinke, EN flinc = Q. filinke; {kele >>} tele “silver” beside {kelepe >>} telepe (PE21/71-72 and note #11).

The revision of kele/kelepe >> tele/telepe reflects the change of ON >> OT; the usual Telerin word for “silver” was T. telpe (telep-) and in Sindarin S. celeb (UT/266). At the time this note was written, Noldorin and Telerin were still unrelated languages, but the revision indicates Tolkien believed the sound change might apply to either language. Assuming the phonetic change applied to both Sindarin and Telerin, it seems logical that the change occurred in the shared Ancient Telerin period of these languages, but the evidence itself is inconclusive.

There are a number of clear examples of this development in Noldorin: ᴹ√PEL(ES) > ON. pele > N. pêl, ᴹQ. peler “fenced field” (Ety/PEL(ES)); ᴹ√NÉL-EK > ON. nele > N. nêl, ᴹQ. nelet “tooth” (Ety/NÉL-EK); ᴹ√skyapat- > ON. skhapa > N. habad, ᴹQ. hyapat “shoe” (Ety/SKYAP). As the last example illustrates, the “lost” final consonant was often restored by analogy with plural: in this case N. hebeid < ON. skhapati, resulting in N. habad rather than expected **hâb. The original final consonant would be retained in plurals because the plural suffix -ī kept it from becoming final; later comparison with the plural could then lead to the final consonant’s restoration in the singular as well. Compare also the variants N. nêl vs. N. neleg “tooth”, the latter a restoration based on the plural.

Restoration of final p, t, k would have been common, but lost s was not restored, because intervocalic [s] became [h] in Old Sindarin and then later this [h] was lost, even in plural forms. There was therefore no consonant to restore from the plural. This sometimes resulted in some irregular plurals, as described in the next example.

There aren’t many clear examples of this final consonant loss in Sindarin, but one possibility is S. thôl vs. Q. solos “helmet” (PE17/188), probably from primitive [tʰolos]. This Sindarin noun had an irregular plural thely, and the likely explanation is that the resulting intervocalic [h] in the plural form preserved the second o for enough time that it was reflected in the plural: [tʰolosī] > [θolohi] > [θoluhi] > [θœlyhi] > [θely].

It is unclear whether or not the rule could apply to monosyllables as well as polysyllables; there are no attested examples. Given the large number of primitive monosyllables to which this rule might apply but did not, it seems safer to assume the change occurred only in polysyllables. See, for example, ᴹ✶d’rāk > N. draug “wolf” instead of **drau (Ety/DARÁK).

Conceptual Development: Determining the final consonants of primitive words is quite difficult in the earliest stages of Tolkien’s languages, so it is anyone’s guess whether this rule also applied to Gnomish.

My rationale for including this phonetic rule in the Ancient Telerin period is entirely due to the note from Common Eldarin Noun Structure given above (PE21/71-72), and I could easily be wrong. David Salo proposed these phonetic rules in Gateway to Sindarin before the publication of PE21, and attributed them to the Old Sindarin period instead (GS/§4.53, 4.72).

See AT. final voiceless stops and [s] vanished in polysyllables for discussion.