I took the liberty to slightly modify Ed's example to get a better grip on these issues. (It is also better to disable debugging and crank up the priority to compete less with other processes).
It is important to vary the iteration counts of the outer and inner loop to get a better picture. With an outer loop of 1000 and an inner loop of 10, the property versions are about 25x slower than the wire version. This can get significant! It is like going back to a 200MHz Pentium 1
😉The timing differences are lower of the array sizes increase.
One puzzling thing is the fact that the local variable version seems to be always faster than the wired version. Very interesting!
(Use: Select iteration counts, press "Test!", wait for results)