If I read Malzahn’s arguments regarding the dual endings correctly, she suggests that their exceptional behavior can be attributed to the frequency of pre-pausal forms in numerals and vocatives, and that in the latter case prosodic factors may have played a role. 

I’m not convinced, however, that this will explain the lack of sandhi; if anything, a short final vowel might be expected to undergo contraction even more readily than a long final vowel. Moreover, there are many forms in short final vowel (e.g. the vocative in -a of a-stems) that do not exhibit exceptional sandhi behavior. So the situation still looks murky to me.

Best wishes,

Hans Henrich