"Substantial", maybe. I don't recall any major papers this year about efficiency in NeurIPS, ICLR, etc.

Nevertheless I think my point on "the wider Deep Learning community" not focusing on efficiency is correct.