The vector average offset would produce a homogenous output from multiple inputs. It would also blur the subtle distinction(s) between distinct input queries that could be important.
But this approach of vector averaging would use less resources. So it’s a quality/compute trade IMO.