Speaker variation is sometimes documented in the Australian numeral systems, but not systematically so. For Bardi, speakers differ greatly in the extent to which they accept numeralsbeyond those given in (2). Materials from the 1920s include forms such as gooyarra agal gooyarra agal gooyarra agal gooyarra ’two and two and two and two’ (for 8), but current speakers uniformly described such phrases as ad hoc enumerations which sounded contrived and ungrammatical. However, the presence of such formations in earlier materials – which date from a time when Bardi was still used in daily conversation – may suggest that the system has contracted as speakers’ knowledge of English has increased.
I note in passing that the patterns in that first paper (that is, limits on numeral systems lining up areally, with Australian and Khoisan languages not having many numerals, other African languages varying but tending toward low limits, and Asian languages having high limits) look like they line up well with the IQ data I’ve seen, although South America is mildly surprising.
A semi-relevant paper: http://www.academia.edu/1917177/On_numeral_complexity_in_hunter-gatherer_languages
A footnote:
Also: https://numberwarrior.wordpress.com/2010/07/30/is-one-two-many-a-myth/ -- the urapon/ukasar thing suggests that you might be right, but that looks like the same structure as in Bardi.
I note in passing that the patterns in that first paper (that is, limits on numeral systems lining up areally, with Australian and Khoisan languages not having many numerals, other African languages varying but tending toward low limits, and Asian languages having high limits) look like they line up well with the IQ data I’ve seen, although South America is mildly surprising.