Well, most of the time foreach and map do different things, so I don't know about that. Maybe for (@in) { push @out, op($_); } is faster than @out = map { op($_) } @in;, but I think what you read might have been about the use of map as a substitute for for, as in ; map { $_ *= 2 } @list; instead of ; $_ *= 2 for @list;. You might read that it's bad practice because map builds a new list which is thrown away (people will tell you not to use map in void context). I think it is bad practice because map is supposed to turn a list into another, so the intent is clearer in the second case.

And to answer your question, since for can't be a sub expression (you shouldn't use the return value of a for loop), but the looping part is the inner part of the logic (the outer part being to choose between the cache and a new computation), to use the for loop, you need to extract the if/else logic out.  $sum = $cache{$inner} //= sum map { $_->{value} } @{ $foo->{$inner} }{@keys}; becomes:

unless (exists $cache{$inner})
{
  $cache{$inner} = 0;
  $cache{$inner} += $_->{value} for @{ $foo->{$inner} }{@keys};

  # This might be a little faster, because perl doesn't have to build 
+the temporary list @{ $foo->{$inner} }{@keys}
  # $cache{$inner} += $foo->{$inner}{$_}{value} for @keys;  
}
$sum = $cache{$inner};
[download]

i have tried that and $cache{$inner} += $_->{value} for @{ $foo->{$inner} }{@keys}; is slightly slower than $cache{$inner} += $foo->{$inner}{$_}{value} for @keys; but both are slower (just about 1s) than map, so i will stay with your first version !

now my code run in 12s instead of 30s with my version, i would not have hoped for such optimization, Perl and YOU are magic ! ;)