use strict;
use warnings;
use Data::Dump 'dump';

my $INTERVAL = 500;

# Don't know how you get your indices, so:
my ( $x, $y, $z ) = ( 1, 2, 3 );
my @indices = ( $x, $y, $z );

my %orig = (
  I => {
    $x => [ 4, 10, 200, 6000, 7000, 7100 ],
    $y => [ 75, 350, 5900, 6402, 7110 ],
    $z => [ 0, 95, 5990, 6450 ] },
  II => {
    $x => [ 0, 20, 501, 1000 ],
    $y => [ 22, 23, 24 ],
    $z => [ 22, 23, 500 ] },
);

my %outhash;

foreach my $key ( keys %orig ) {
  my ( @merged, @graded, @prev );

  foreach my $a_ref ( @indices ) {
    push @merged, map { $_ . '.' . $a_ref } @{ $orig{$key}{$a_ref} };
  }
  @merged = sort { $a <=> $b } @merged;

  foreach my $i ( 0 .. $#merged ) {
    my @cluster;
    foreach my $j ( $i .. $#merged ) {
      last if $merged[$j] - $merged[$i] > $INTERVAL;
      push @cluster, $merged[$j];
    }
    if ( not subset_of( \@prev, \@cluster ) ) {
      push @graded, [ @cluster ];
    }
    @prev = @cluster;
  }

  foreach my $a_ref ( @graded ) {
    my %HoA;
    for ( @{ $a_ref } ) {
      my ( $int, $idx ) = split /\./;
      push @{ $HoA{$idx} }, $int;
    }
    push @{ $outhash{$key} }, [ %HoA ];
  }
}

dump( %outhash );

sub subset_of {
  my %first = map { $_ => undef } @{ $_[0] };
  my @second = @{ $_[1] };
  for ( @second ) {
    return 0 unless exists $first{$_};
  }
  return 1;
}