Gaze goes a bit beyond merely focusing on the source of sounds. I think it's fair to say that not everyone is going to do this all the time. Context and the person(s) involved matter:
Decades of research into the use of gaze in natural interactions [...] shows how gaze can signal attentiveness, competence and social dominance. This means that gaze can be a positive, neutral or negative cue, depending on the context. Direct gaze can be used to regulate conversation shifts [...] and to signal social interest [...]. Prolonged gaze or staring leads to avoidance behaviours [...]
So really interesting or dumb questions are more likely to get a head turn. And the participants doing the head turns are more likely to be those really interested or at least not socially shy. Also, long exchanges with the podium speaker are more likely to get that either because the exchange could be really interesting or perceived as really boorish. In that latter case a long gaze is a form of peer pressure to shut up and let the podium speaker continue; or maybe just let the rest of the students leave the amphitheater if this is an end-of-class question.
Since you're from Japan, I tried to find if there's a cultural difference with Europe or US on this issue specifically (gazing at someone asking questions in class or other large audience)... but I couldn't find anything particular insofar. In general, it is known that gaze behavior is culture-related to some extent, e.g. Japanese tend to gaze less at someone (compared to Canadians) when answering a question.
Another interesting issue would be how many students initiate this type of gaze and how many are simply respond (follow) the gaze of others... The two are distinct processes:
Whereas first elements of RJA [Responding Joint Attention] are already evident at 6 months of age, IJA [Initiating Joint Attention] does not emerge before the second year of life (Mundy and Newell, 2007; Mundy et al., 2007). Chimpanzees followed the experimenters gaze on a frequent basis but did not try to initiate JA (Tomasello and Carpenter, 2005). Interestingly, differential development of both RJA and IJA can be observed in brain systems from childhood to adulthood (Oberwelland et al., 2016), as well as during atypical development in disorders such as autism (Oberwelland et al., 2017). In autism, IJA is typically more impaired than RJA and emerges much later than in typical development (Mundy, 2003). These empirical findings clearly point toward separate underlying cognitive systems of RJA and IJA (Mundy and Newell, 2007).