Philomen Probert (Wolfson College, Oxford) writes that
"[A] nominative/accusative dual ending in ω always has an acute, never a circumflex, if accented on the final syllable, regardless of contraction:
νόω > νώ (not *νῶ); ὀστέω > ὀστώ (not *ὀστῶ)
(Probert 2003, §112). This is a synchronic observation.
Other similar exceptions (to the normal rules of contracted forms) she mentions are:
- adjectives of material in -οῦς from -εος (§122);
- contracted nouns and adjectives in -ους from -οος (§121);
- feminine nouns in -ώ (§127).
As for why, I'm not quite sure yet. We “inherited” this convention from the ancient grammarians (Apollonius and his son Herodian, John Philoponus, Theodosius of Alexandria, George Choiroboskos) - see Chandler 1881 §563 for further details, with relevant original passages in Greek on this question.
NB: A lot of Greek grammars are secondary sources and are rather compilatory. In other words, quite often they merely repeat the so called "rules" from the earlier books. For instance, it seems that Kühner–Blass base their observation on Göttling 1835 Allgemeine Lehre vom Akzent der griechischen Sprache (Chandler 1881 basically repeats him too, adding only one other quotation).
To sum up, this is a “rule” passed upon us from the ancient grammarians. In present-day linguistic research it is a thorny issue and is usually mentioned as an exception (e.g. Kiparsky , Probert 2003).
Of course, we could question the accuracy of linguistic description of the Alexandrian grammarians; that being said, it seems inscriptional data confirms the dual oxytone in some dual forms (cf. Monatanari "dual κανώ in inscr.")
On the other hand, it could be explained internally; e.g. Steriade 1988 explains the nominative dual oxytone - to use her terminology, accent on a desinential mora - with morphological factors, in other words, with analogy; cf. Papanastassiou 2013 "[t]he accent of the nominative is kept in all cases [in the Attic declension - Alex B.]. The genitive and dative are oxytone when the final syllable is accented, even if they result from contraction."
If I understand Kurylowicz 1958 L'accentuation des langues indo-européennes correctly, he believes the oxytone of the dual nom-acc marked can be explained analogically. Here's the relevant quote in the original:
"Cette constatation est importante pur le problème de l'ancienne intonation des duel en -ω, -ᾱ. Il s'agit là évidemment d'un phénomène d'ordre sémantique et morphologique. Le manque d'intonation, c.-à-d. l'intonation aiguë, dans ὀφθαλμώ devient compréhensible si l'on suppose qu'au moment de l'introduction du circonflexe dans les désinences -ῆς, -ῇ, -ῷ, -ῶν, le duel ne jouissait plus des droits d'une forme appartenant au paradigme normal, mais était sémantiquement subordonné au pluriel et en quelque sort dérivé de lui (-ώ, -ᾱ́ dérivés < -οί, -αί). C'est un indice de la genèse relativement tardive du système de l'accentuation grecque" (Kurylowicz 1958: 128-129).
-if anyone is willing to post its English translation, be my guest; too tired typing all the French and Greek forms on an English keyboard -
also cf. Bubenik 1983 "The Nom/Acc Dual did not develop an analogical circumflex because the dual is rather a derivational category" [emphasis mine - Alex B.].
Or we could explain it externally - in other words, we could trace it to the PIE dual marker *o-h1(e) and its development in the IE languages (cf. Lithuanian and Vedic dual forms); cf. φῶς < φάος (internal) and φώς (external) (Krasukhin 2004: 88).
Also, I found it interesting that Alonso Deniz 2013 argues that contraction is generally avoided in disyllables as opposed to grammatical disyllables, which may contract.
A note on κανοῦν (κάνεον). Smyth and Messing write that it got its accent from its genitive and dative forms (§236b).