With masculine and neuter nouns, два and три take the genitive singular. With feminine nouns it is at least usually taught that две and три also take the genitive singular. But how do we know? Given that this form of a feminine noun is almost always the same as the nominative plural, can we not with equal justification say that these numerals take the nominative plural? Or might we not say that they take the form that expresses both of those functions?
But how do we know?
Historically, Russian cases are defined by the questions the nominals put into each case answer.
This might not the best way to define a case, because using this definition you can't tell partitive from genitive as in чая vs чаю, or locative from prepositional as in о лесе vs в лесу, but it's how it works in traditional grammars which trace back to Meletius' Church Slavonic grammar.
However, this definition is good enough to tell the two cases apart in your case. You can't ask in Russian две кто? or две что?, you'd have to ask две кого? or две чего? which undoubtedly puts the corresponding word into genitive or its varieties (or which nominative is not).
Given that this form of a feminine noun is always the same as the nominative plural, can we not with equal justification say that these numerals take the nominative plural?
The feminine nominals are not always the same as nominative plural as in genitive singular.
The paradigm differs for all adjectives and ordinal numerals:
- красивые (f. nom. pl.) / красивой (f. gen. sg.)
- первые (f. nom. pl.) / первой (f. gen. sg.)
, for nouns with mobile stress:
- ре́ки (nom. pl.) / реки́ (gen. sg.)
- во́ды (nom. pl.) / воды́ (gen. sg.)
, for nouns with a mobile vowel in the root:
- сёстры (nom. pl.) / сестры (gen. sg.)
- зори (nom. pl.) / зари (gen. sg.)
, for ad-hoc declension paradigms resulted from the gender shift:
- гроздья (nom. pl.) / грозди (gen. sg.)
This said, numerals in две, три and четыре undoubtedly govern singular genitive even for feminine nominals.
I'd argue that the case of сестра is a proof that the Genitive feminine form is singular and not plural. Other such words are башка (if we accept its Nom. pl. as бошки, Gen. - башки) and весна (вёсны vs весны respectively), зима (stress variation: зИмы vs зимЫ respectively), беда (stress variation: бЕды vs бедЫ respectively).
We also may consider two other genders which clearly provide singular forms and i personally as a native speaker (which isn't of course any proof) perceive the feminine flexion as of a singular form and because with numerals you expect some sort of oblique case.
In my opinion the forms of feminine Nom. pl and Gen. sing (in this particular case) are simply homonyms, they are spelled and often sound alike but function differently.
But maybe historical philology can provide more conclusive clues.