Well, it can be made a little more compact. The following doesn't bother validating data, although such validation is fairly trivial:

use strict;
use warnings;

my @res = analyze ();

print "$_->{f1},$_->{f2},$_->{f3}\n" for @res;


sub analyze {
    # get the data as an array of hashes just like DBI would return
    my @data = map {
        chomp;
        my @f = split /,/;
        {f1 => $f[0], f2 => $f[1], f3 => $f[2]}
    } <DATA>;

    my %lists;

    # Build coloured lists
    push @{$lists{$data[$_]{f2}}}, $_ for 0 .. $#data;

    # Find the spans
    my $redIndex = $lists{Red}[0];
    my @priorWhites = grep {$_ < $redIndex} @{$lists{White}};
    my $firstWhite = $priorWhites[-4];
    my @priorBlues = grep {$_ < $firstWhite} @{$lists{Blue}};
    my @postWhites = grep {$_ > $redIndex} @{$lists{White}};
    my $lastWhite = $postWhites[3];
    my @postBlues = grep {$_ > $lastWhite} @{$lists{Blue}};

    # Build the output list
    my @indexes = ($priorBlues[-1], @priorWhites[-4 .. -1], $redIndex,
        @postWhites[0 .. 3], $postBlues[0]);
    return @data[@indexes];
}


__DATA__
...
[download]

generates output per OP's sample given OP's data.