my %seen;
while (<>) {
  chomp;
  my @fields = split('\|', $_);
  if (my $record = $seen{$fields[0]}) {
    ... append @fields to $record...
  } else {
    $seen{$fields[0]} = ...some data structure...
  }
}
for my $k (keys %seen) {
  my $record = $seen{$k};
  # process $record
}
[download]

Is each record one long line or is it 4 lines consisting of a | separated line, two alphabet lines and a blank line?

Also, do you know if your input is sorted? Even though perl can handle reading in 10K lines into memory, we can optimize the code if we know that the input is already sorted by id.

For the following I will assume that each record is 4 lines:

my %seen;
$/ = "\n\n";
while (<>) {
  chomp;
  my ($idline, @letters) = split(/\n/, $_);
  my @fields = split('\|', $_);
  my $record = $seen{$fields[0]};
  unless ($record) {
    $seen{$fields[0]} = $record
       = { idline => $idline, letters => [ @letters ] };
  } else {
    push(@{$record->{letters}}, @letters);
  }
}
for keys (%seen) {
  print join("\n", $seen->{$_}->{idline}, @{$seen->{$_}->{letters}}, "
+\n"), "\n";
}
[download]