AQ. nasals became voiceless stops before aspirates; [mpʰ|ntʰ|ŋkʰ] > [ppʰ|ttʰ|kkʰ]
Where a nasal preceded an aspirate in Ancient Quenya, the nasal unvoiced and assimilated to the aspirate, producing an aspirate pair which then produced a pair of voiceless stops. Thus mph, nth, ŋkh > pph, tth, kkh > pp, tt, kk. Tolkien described this change in both the Outline of Phonetic Development [OP1] from the 1930s (PE19/44) and the Outline of Phonology [OP2] from the 1950s (PE19/89).
This change was resistant to analogical leveling, so that the stop-pair tended to survive even in past tenses. The example Tolkien gave was rappe past tense of raf- “seize” (1930s) or “wave” (1950s) from the root √RAPH. However, after the TQ sound change þ > s, the past tenses of verbs originally ending in the aspirate th tended to be reformed, such as panse (replacing †patte) as the past tense of pas- “to smooth” from the root √PATH (PE19/88-89).
Conceptual Development: There are no signs of mf or nch at any conceptual stage of Quenya, but aspirates were not a part of Tolkien conception of early primitive Elvish. Based on forms like ᴱQ. tempe “hate” < ᴱ√TEFE and ᴱQ. ranta “bridge” < ᴱ√RAÞA, it seems that voiceless spirants became stops after nasals in Early Qenya. In the 1930s and 1950s it was also the case that spirants became stops after nasals where they came into contact with each other later in Quenya’s phonetic development (PE19/44, 89), but this sound change was after mph, nth, ŋkh > pph, tth, kkh in both OP1 and OP2.
Even in his later writings, Tolkien may not have been 100% committed to the stopping of nasals before aspirates. In some rejected notes on irregular verbs from the 1940s, Tolkien explored alternate phonetic developments next to root ᴹ√EPH that seem to show the nasal being retained in past tenses:
EPH-. (especially from water, opp[osite] of “dive”) emerge. épha it emerges. empe. emphe. [nearby in the margin:] ephe, emphe. ethe, enthe. ehe, ēhe. eñghe; enche (PE22/127 note #127).
Perhaps in the rejected note above he imagined this was the normal development in all cases, and nasal stopping did not occur. However, elsewhere in the same document on verbs he gave a past tense paradigm matching that of OP1 and OP2:
3) ph, th, kh. Not many strong verbs with these medials survive. Where they do they preserve the historic forms with pp etc. < mph etc. with occasional forms derived from phn > pt, khn > kt, ht (ñkh, khn [???]). RAPH, snatch: Q raphe, rafe: pa.t. rappe (PE22/102).
This section was not rejected, unlike the text containing ᴹ√EPH. Thus it seems the phonetic developments next to ᴹ√EPH were a transient idea.