Beni-Snous: Two unrelated phonetic forms for every noun?

I got flabberghasted recently by a casual statement in Destaing (1907:212)'s grammar of the Berber dialect of Beni Snous in western Algeria (near Tlemcen). I nearly missed it as I skimmed it; see if you can spot it. (The translation is mine, as are the bits in brackets.) All the numerals above 1 are from Arabic here, but that's nothing surprising - the same is true in Tarifit, and few Berber varieties have retained the numbers above 3.

"The numbers from 2 to 9 inclusive are followed by the Berber noun in the plural [eg]:

two men ..... θnāịẹ́n ịírgǟzĕn
six women ... sttá n tsénnạ̄n
[...]

From "10" to "19" inclusive, the number is followed by the Arabic singular substantive:

eleven women ... aḥdăɛâš ĕrmra (Algerian Arabic mṛa "woman" مرة; contrast Beni Snous Berber θä́mĕṭṭūθ "woman")
fifteen cows ... ḫamstaɛâš ĕrbégra (Algerian Arabic bəgṛa "cow" بڨرة)
sixteen mares ... sttɛâš ĕrɛấuda (Algerian Arabic `əwda "mare" عودة; contrast Beni Snous Berber θáimārθ "mare")

After the number nouns "twenty, thirty, forty" etc., one uses the Arabic substantive[...]

twenty women ... ɛašrîn ĕmra
fifty mules ... ḫamsîn beγla (Algerian Arabic bəγla بغلة "mule")

a thousand rams: âlĕf kebš (Algerian Arabic kəbš كبش "ram"; contrast Beni Snous Berber išérri "ram")"

If I thought it were remotely possible for Destaing's claim to be true of counting every noun in the language - rather than, say, just the six nouns he gives appropriate examples for - I would be putting together an application to head out to Tlemcen instead of making this posting. (I might still do that anyway some time, mind you.) But for rather a lot of minority languages, all or nearly all speakers are bilingual. And if all speakers are bilingual, what in principle is there to prevent the grammar from containing a rule like this?

So I ask: have you ever come across anything similar elsewhere?