Yes, of course. Your benchmark is even more appropriate to show the overhead, which consists in creating, populating and destroying a hash (call with hashref) vs. populating a hash already allocated on the subroutine's pad (call with list).

Might be small, but might sum up. Call it a micro-de-optimization, but I see no value in making perl slower for a questionable gain and a loss of readability.