The idea was to be able to use map to build the return list. Obviously that makes little sense in void context, which is all your for proposition will be able to provide.
undefs are actually less wasteful than actual values - my @x = (undef) x 10,000,000; consumes 130MB for me, my @x = ('a'x10) x 10,000,000; nearly hits the 500MB mark.
Of course, if you're not in void context and actually intent on returning the resulting list from processing a 10,000,000 key hash, you'll have to be able to fit that in memory anyway. Since you're throwing around big chunks of data, memory can't be a huge concern, otherwise you'd be walking the hash manually and chewing the bits more carefully.
You can't have your cake and eat it - you can't be using an iterator when you're concerned about memory usage.