Very interesting info, thank you. I often wonder about such efficiency when writing perl but it doesn't usually matter too much. I do write postfix loops because they're pretty and one-line friendly (I ❤ print$_ for@_) but not when they're nested because I tend to forget about them. If they're really much faster I could use comments to remind myself. Anyway your version is slightly faster (~0.750) but the data has whitespace and when using map to remove that it gets even slower than mine. What's going on?

time perl -MAlgorithm::Combinatorics=:all -wle'
my $i=variations_with_repetition(["a".."z"],$ARGV[0]); 
my @x; push @x, map {s/ //gr} qq[@$_] while $_=$i->next; 
print scalar @x' 4
[download]

real 0m1.050s

Profile:

perl -d:NYTProf -MAlgorithm::Combinatorics=:all -wle'
my $i=variations_with_repetition(["a".."z"],$ARGV[0]); 
my @x; push @x, map {s/ //gr} qq[@$_] while $_=$i->next; 
print scalar @x' 4; nytprofhtml --open
[download]