Both scale the same for insertions (amortised O(1))

Care to demonstrate how insertions into an array amortise to O(1)?

Both scale the same for fetches (amortised O(1)).

A fetch should take a little longer with a hash, but I suspect this wouldn't be the deciding factor in most cases.

You wouldn't notice if your programs ran 5 times slower?

perl -MTime::HiRes=time 
-E"@a=1..1e6;$t=time; ++$a[$_] for 1..1e6; printf qq[%f\n], time()-$t"
0.146912

perl -MTime::HiRes=time 
-E"%h=map{($_)x2}1..1e6;$t=time; ++$h{$_} for 1..1e6; printf qq[%f\n],
+ time()-$t"
0.525000
[download]

Or an order of magnitude more slowly?

perl -MTime::HiRes=time 
-E"$t=time;my @a; $a[$_]=1 for 1..1e6; printf qq[%f\n], time()-$t"
0.220181

perl -MTime::HiRes=time 
-E"$t=time;my %h; $h{$_}=1 for 1..1e6; printf qq[%f\n], time()-$t"
1.805000
[download]

Concealing reality behind grand sounding theoretical BS is all the worse as you obviously know better.