Use Text::CVS to do the heavy lifting:

#!/usr/bin/perl -w 
use strict;  
use warnings;
use Text::CSV;

my $csv = Text::CSV->new();

while (defined (my $line = <DATA>)) {
    next if ! $csv->parse ($line);
    
    my @fields = $csv->fields();
    
    $fields[2] = join ' ', reverse split /,\s*/, $fields[2];
    $csv->print (*STDOUT, \@fields);
    print "\n";
}


__DATA__
-###.####,##.###,"Doe, John & Jane","### Main St","Town, State 13370",
+,,
-###.####,##.###,"Doe, John","### Main St","Town, State 13370",,,
[download]

Prints:

-###.####,##.###,"John & Jane Doe","### Main St","Town, State 13370",,
+,
-###.####,##.###,"John Doe","### Main St","Town, State 13370",,,
[download]

Note that there was a missing " in the second line of the sample data which I presume was a typo. If it wasn't a typo you've got a bigger problem - your data is corrupt!