This produces the required results for the testcases provided reasonably efficiently. Generalising the implementation is left as an exercise.

Basically, you replace spaces with nulls, OR the n-1 shorter strings together and then XOR the result with the longest string. You then count the number of nulls in the result.

The definition of "shorter" and "longest in that description is fuzzy :).

#! perl -slw
use strict;

my $s1 = 'GATTACGAGTGGCGCTCGTGTAACGGCA';
my $s2 = 'GATTACG    GCGCTCG   AACGGCA';
my $masked = $s1 ^ $s2;
print scalar $masked =~ tr[\0][0];

my $s3 = 'GATTACGAGTGGCGCTCGTGTAACGGCA';
my $s4 = 'GATTACG                     ';
my $s5 = '  TTACGAG       CGTGTAA     ';
tr[ ][\0] for $s3, $s4, $s5;

$masked = $s3 ^ ( $s4 | $s5 );
print scalar $masked =~ tr[\0][0];

my $s6 = '             GCTCGTG        ';
my $s7 = 'GATTACGAGTGGCGCTCGTGTAACGGCA';
my $s8 = '   TACGAGT                  ';
my $s9 = '        GTGGCGC             ';

tr[ ][\0] for $s6, $s7, $s8, $s9;

$masked = $s7 ^ ( $s6 | $s8 | $s9 );
print scalar $masked =~ tr[\0][0];

__END__
P:\test>junk2
21
16
17
[download]