Surprise...

[Brian Milby]

I expanded my test stack and ran it on the versions I had available.
Here's a couple of tests from each version:

macOS Sierra (10.12.6) on an iMac (Retina 5K, 27-inch, Late 2014) 4 GHz i7
9.0.1-rc-1 Business / 8.1.9 Business / 7.1.4 Indy / 6.7.11 Indy

34 10 11 10 34 5
LC9 item: 4562 / array: 12660 / 0.360348
LC8 item: 6175 / array: 13111 / 0.470979
LC7 item: 10994 / array: 11554 / 0.951532
LC6 item: 1278 / array: 4681 / 0.273019

340 10 11 10 34 500
LC9 item: 430179 / array: 125709 / 3.422022
LC8 item: 449724 / array: 131206 / 3.427618
LC7 item: 2473784 / array: 114818 / 21.545263
LC6 item: 258495 / array: 45592 / 5.669745

Notice that even in LC6, for this particular test (the original one), item
is faster (and by a bigger margin).  I don't think it is as much about the
speed of arrays per se, but knowing the data set and using the right
approach to get the optimal performance.  So, it is probably a reasonable
approach to use arrays first.  If you need to optimize for speed, this is
an area to consider.

Thanks,
Brian