If a hiring decision is influenced to 0.01% by speech and accent and 99.99% by whatever you deem reasonable criteria, then what do you learn from filtering out the latter?
And of course, decisions like this need not have linear behavior in reality either. A gross mismatch between different indicators could result in significantly worse hiring manager opinions than the (weighted) sum of the individual aspects.
Imagine that a candidate shows moderate (but sufficient) aptitude but then tries to disguise it with "high brow" rhetoric. In isolation using such speech may be considered a positive, but in this case it will be probably be to their detriment.
Given that one small element of interview presentation has been shown to involve irrational bias, perhaps it's more reasonable to assume that most others also involve unconscious or explained-away bias? So perhaps hiring decisions are influenced 0.01% by speech and accent, 64.99% by other class, race and gender markers, and 35% by reasonable criteria such as competence? Why wouldn't this be your base case assumption?
No such thing was shown in this study. The study makes no attempt to measure E(skills | social class). For all we know, the hiring managers provided an unbiased (i.e. the opposite of "irrational bias") estimate of the expected skills conditioned on the social class of the applicant, absent any other information.
That seems like an extraordinary claim unless you have some evidentiary base for it.