This works for me. There might be some glitches in the input format not shown in the sample, though.

#!/usr/bin/perl
use warnings;
use strict;

use Data::Dumper;

<>;                                                   # Skip the first
+ line.

my %hash;
my ($cell, $cellr);

while (my $header = <>) {
    if ($header =~ /\b/) {

        if ($header =~ /^CELL\s*$/) {
            $cell = <>;
            chomp $cell;
            next;
        }

        my @pos;
        my $flip;
        while ($header =~ /(\b|$)/g) {
            push @pos, $-[0] if ++$flip % 2;          # Remember the p
+osition where a field starts.
        }

        chomp (my $value_line = <>);
        if (length $header <= length $value_line) {
            $pos[-1] = length $value_line;            # Do not clip th
+e value line if longer then header.
        } else {
            $value_line .= ' ' x (length($header));   # Do not die if 
+the last fields are empty.
        }

        my @values;
        my @fields = split /\s+/, $header;
        for my $i (0 .. $#fields) {
            push @values, substr $value_line, $pos[$i], $pos[$i+1] - $
+pos[$i];
        }
        s/^ +| +$//g for @values;

        if ($fields[0] eq 'CELLR') {
            $cellr = shift @values;
            shift @fields;
        }

        $hash{$cell}{$cellr}{$_} = shift @values for @fields;
    }
}


print Dumper \%hash;
[download]

Updated: Keep the whole structure in one hash.