use strict;
use warnings;
my %hash;
while (my $line=<STDIN>){
  foreach my $word (  split( /[^a-zA-Z]+/ , $line)   ){
    my $len = length($word);
    $hash{lc $word}++ if 2<=$len && $len < 5;
  }
}
print $_."\n" for sort keys %hash;
[download]

• be sure to 'use strict' and 'use warnings' .. (thus need 'my $line' and 'my $word')
• note that lowercasing isn't done til last.
• replaced m/^.?$/ regex with length (i assume that's what it was doing -- length should be faster, and clearer, than a regex)
• removed the !~m/(\w)\1\1\1\1/ and replaced with a length check for speed and clarity. (don't want 5+ letter words, right?) Update: as PerlMouse pointed out I misread the regex--it's excluding words with a letter repeated 5 or more times
• removed the s/[^a-z ]+/ /g; and replaced it implicitly with the regex in the split()... now have it split on non-leters, which all go away and you're left with the words of just letters. Not having the substitution should help a lot speed-wise.