Certainly in the circles I'm from in the UK, less/fewer is very much used as a signal. I don't think I could use the 'wrong' one without getting corrected if the audience is sufficiently large.

Question: In casual conversation, does the proportion of the time I am corrected increase with the number of people as if they each corrected as iid Bernoulli random variables? (i.e. if I get corrected 1/2 of the time with one other person, then it's 3/4 of the time with 2, 7/8ths of the time with 3 etc.)

I suspect that I would be corrected more often than that model predicts in larger groups, because there are more people to signal status to.