A vowel may become "unstressable" for
various reasons. Your "Strange Rule" is less strange than many I have seen.
Imagine a hypothetical language in which the folowing principles
(1) As few underlying vowels as possible should be realised.
(2) Underlying /a/ is the preferred target of elision.
(3) Stress must be word-initial.
There will be other constraints as well, defining permissible syllable
structure and minimal word shapes (e.g. "The stressed syllable must contain a
vowel") and therefore preventing the deletion of some vowels, but I'm interested
here only in the interaction of (1), (2) and (3). An underlying sequence like
/tiruki/ will be realised as ['tirki] (if, say, final clusters like
[-rk] are not permitted), but /taruki/ will be realised as ['truki], since
*['tarki] would violate (2) while satisfying (1) and (3) to the same extent as
['truki]. In cases like /taraki/, ['tarki] and ['traki] are equally well-formed
and the resulting tie may be resolved by constraints not discussed here (e.g. if
the language avoids complex onsets, ['tarki] may turn out to be preferable to
['traki]). I wonder what happens to such words in your model.
Such a state of affairs could arise in a language in which stress was
originally initial, and in which vowels underwent reduction depending not only
on the position of stress but also on their quality. The impression one gets
after the system has been implemented is that initial stress has been shifted to
the second syllable in some words. Actually, the stress remains
initial, only its original carrier disappears so it finds a new one. A
similar cases is Proto-Slavic, in which short *i and *u underwent gradual
reduction via lax *I and *U (the "yers") to zero. We could say that in a
typical early Slavic language a word contained as few yers as possible. But
in words containing _only_ such vowels, e.g. *sUnU 'sleep' < *supnu <
*supnos there was a conflict between this principle and pronounceability
requirements (*sn was ruled out). Since final weak vowels were
_preferentially_ dropped, the optimal compromise was [sUn] (Russian son, Polish
sen, Serbian/Croatian san, etc., with the second yer lost but the
first granted full-vowel status). In inflected forms, however, the second
syllable usually contained a full vowel and then it was the first yer that was
weakened and lost: *supnoi (Loc.) > *sUne^ > sne^.