Primitive Elvish allows for tri-consonantal combinations only when those combinations end in a y or w. In many of those clusters, the final y, w developed into i, u, as described in the Outline of Phonetic Development (OP1) from the 1940s and the Outline of Phonology (OP2) from the 1950s:

After a consonant group they [y, w] became syllabic i, u (with weak off-glide ı̯, u̯) — except that asyllabic y was retained and favoured after k, g (> t, d), t, d; and w after k, g. In these cases a diphthong or a long vowel could also precede: so mátya, maitya beside mantya, but kalt-wa > kaltua (OP1: PE19/48).

Since th, kh > t, k after r, l in the infrequent groups rthy, lthy; rkhy, lkhy; rkhw, lkhw: thy, khy > ty (for which see under Voiceless Stops); and khw > kw (qu). In other combinations y, w became syllabic: thus lphy > lpi; rphy > rpi (phw did not occur) (OP2: PE19/90).

After consonant groups y, w became syllabic i, u, except that y remained after t, d (either original or derived in Quenya from k, g); while w remained after k, g. Before retained ty or tw a long vowel or diphthong (durational equivalents of short vowel + consonant) could also stand. Thus possible in Q were mátya, maitya, mantya (dissyllabic); whereas mátwa, maitwa, mantwa > mătwa (or matua), maitua, mantua (OP2: PE19/100) [note #153: additional valid forms saikwe, laukwa were written nearby].

The net result is that all tri-consonantal clusters in Quenya ended in ty, dy, kw, gw, the first two of these possibly developing from ky, gy after velars dentalized before [j]. Tolkien listed the valid clusters as:

• hty, sty, skw, nty, ndy, ñkw, ñgw, lty, lkw, rty, rkw (OP1: PE19/38) plus ldy, rdy and (rare) lthy, lhy, lhw, rthy, rhy, rhw (OP1: PE19/51).
• hty, sty, skw, nty, ndy, ñkw, ñgw, lty, lkw, rty, rkw (OP2: PE19/81) plus ldy, rdy (OP2: PE19/100).

As indicated above, it seems Tolkien decided that lthy, lhy, lhw (< lthy, lkhy, lkhw) were no longer possible in the 1950s, as they became instead lty, lty, lkw and similarly for clusters beginning with r, since spirants became stops after [r], [l].

Conceptual Development: In the Primitive Elvish of the 1910s and 1920s, palatals and labialized velars were their own consonant series, so that “clusters” like nty and lkw were probably consonant pairs like [ntʲ] (or [ɲc]) and [ŋkʷ]. Given the behavior of long vowels and diphthongs before ty and kw, the same might be true in their Quenya pronunciation in the 1930s through 1960s (e.g. [tʲ] and [kʷ]); this is an area of some dispute among Tolkienic linguists.