I've run a test that exercises the vm_map structure and measured the execution time and data cache misses. Ten trials with and without the change show that there's no clear time difference, but that this change reliably reduces data cache misses:
without patch:
time 18.945514 dc-misses 741148123
time 18.932776 dc-misses 729875583
time 18.976090 dc-misses 727134615
time 18.944024 dc-misses 722314001
time 18.991595 dc-misses 731146456
time 18.540747 dc-misses 713097975
time 18.599683 dc-misses 716906745
time 18.597823 dc-misses 715262549
time 18.580331 dc-misses 725414998
time 18.532644 dc-misses 713840414
with patch:
time 18.757322 dc-misses 747406950
time 18.933730 dc-misses 754073534
time 18.905024 dc-misses 733378366
time 18.983594 dc-misses 747036044
time 19.42450 dc-misses 735170734
time 18.611044 dc-misses 778169608
time 18.750769 dc-misses 804911585
time 18.580122 dc-misses 788664230
time 18.652336 dc-misses 796562870
time 18.563009 dc-misses 776312074