All credit of this answer goes to sumelic. I just found further support for his hypothesis, of which I was not aware.
This article states:
Bader (1960: 236) remarks that words prefixed by privative in- (<
*en- < *n-) and dis- frequently show vowel reduction (cf. Pultrová
2006: 73, 102-103), such as inimīcus 'unfriendly' (vs. amīcus 'friendly') and difficilis 'difficult' (vs. facilis 'easy'). The list of examples is lengthened by inficētus 'boorish' (vs. infacētus; cf. perfacētus 'very
clever'); īnsipiēns 'unwise' [P1.+], whose finite forms (*insipiō) are
missing in Plautine Latin (cf. insapiēns [Catul.]; persapiēntis 'very
wise' [Cic], nesapius 'ignorant' [Petr.+]); inermis 'unarmed' (vs. arma 'weapons'); iners 'inactive' (vs. ars 'art, skill'); dīmidius 'half (< *dis-; vs. mēdius 'central', cf. rare forms like intermedius, permedius, submedius; see Ernout and Meillet 1985: 393); displiceō 'displease' (vs. placeō 'please', cf. complaceō 'take the fancy', perplaceō 'be thoroughly pleasing'); etc. (but cf. illepidus 'lacking grace', inanimus 'lifeless'). These forms commonly possess a negative sense due to their prefixes. The i-vocalism may thus be phonetically strengthened by a certain emphasis, as Pulgram (1975: 108-109) points out, whereby the
following elements presumably underwent vowel reduction, and were
influenced by the i-vocalism.
References in paragraph:
Bader, Fr. (1960): Apophonie et recomposition dans les composés. RPh 86:
Ernout, A and A. Meillet (1985): Dictionnaire étymologique de la langue latine: histoire des mots. 4th, rev. and enl. ed. Paris: Librairie C. Klincksieck
Pulgram, E. (1975): Latin-Romance Phonology: Prosodies and Metrics. Munich:
Wilhelm Fink Verl
Pultrová, L. (2006): The Vocalism of Latin Medial Syllables. Prague: Univerzita Karlova
v Praze / Nakladatelstvi Karolinum.
Similarly, pages 222-3 of this (old) book state:
Unaccented Latin ă in the posttonic syllable became at first ĕ,
except before l and labials, where it became ŏ. This ĕ became,
perhaps about the end of the third century B.C., ĭ in syllables
not long by position (except when it preceded r), and before ng; while this o became u or the ü-sound, which in most cases passed
into ĭ at the close of the Republican period. Thus the compound
of ab and cado became accĕdo (so spelt by Ennius), then accĭdo;
from in and arma we have the compound inermis; from sub
and rapio first *surropio probably, then surrŭpio (Plaut.), then
surripio', from ex and frango, effringo (see ch, iii. 18). Final Latin ă probably became ĕ, and might be dropped (see ch. iii. 37).