And then there's 1..100_000 that creates an array and 100,000 scalars at compile-time.

Really? Can you demonstrate that?

Because I see no evidence for it:

c:\test>perl -le"++$n for 1..1e2; system qq[tasklist /nh /fi \"pid eq 
+$$\"] "

perl.exe                      7500 Console                    1      5
+,716 K

c:\test>perl -le"++$n for 1..1e3; system qq[tasklist /nh /fi \"pid eq 
+$$\"] "

perl.exe                      3676 Console                    1      5
+,712 K

c:\test>perl -le"++$n for 1..1e4; system qq[tasklist /nh /fi \"pid eq 
+$$\"] "

perl.exe                      6048 Console                    1      5
+,732 K

c:\test>perl -le"++$n for 1..1e5; system qq[tasklist /nh /fi \"pid eq 
+$$\"] "

perl.exe                      5648 Console                    1      5
+,720 K

c:\test>perl -le"++$n for 1..1e6; system qq[tasklist /nh /fi \"pid eq 
+$$\"] "

perl.exe                      8620 Console                    1      5
+,736 K
[download]

Whilst the memory usage varies a little, it sometime goes down a few k when the notional size of this array increases by an order of magnitude?

Conversely, if you explicitly create an array of that size, the memory requirement increases by 100MB as expected:

c:\test>perl -le"@a = 1 .. 1e6; system qq[tasklist /nh /fi \"pid eq $$
+\"] "

perl.exe                      2664 Console                    1    109
+,576 K
[download]