In all cases you don't need to tack on the @ sigil. Your inner array simply needs another index,

foreach my $key (sort keys %hash){
    print $hash{$key}[0];
    print $hash{$key}[1];
    print $hash{$key}[2][0];
    print $hash{$key}[2][1];
}
[download]

my %hash = (
    keya => [a1, b1, [ qw(c1a c1b) ] ],
    keyb => [a2, b2, [ qw(c2a c2b) ] ],
);
foreach my $key (sort keys %hash){
  print $hash{$key}[0];
  print $hash{$key}[1];
  print $hash{$key}[2][0];
  print $hash{$key}[2][1];
}
[download]

note: don't use a comma inside qw() - it expects just spaces.

foreach my $key (sort keys %hash){
  print @{$hash{$key}}[0]; #prints a1 for first hash
  print @{$hash{$key}}[1]; #prints b1 for first hash
  print @{@{$hash{$key}}[2]}[0]; #prints c2a
}
[download]

foreach my $key (sort keys %hash){
  print $hash{$key}->[0];
  print $hash{$key}->[1];
  print $hash{$key}->[2]->[0];
}
[download]

Only that first arrow is necessary. Perl does the rest of the dereferencing without being told:

print $hash{$key}->[2][0];
[download]

----
I wanted to explore how Perl's closures can be manipulated, and ended up creating an object system by accident.
-- Schemer

:(){ :|:&};:

Note: All code is untested, unless otherwise stated

print $hash{"keya"}->[2][1];
[download]

First you get to the first level array by using hash key; Secondly, that second level array ref is the 2nd element of the first level array (0-based); Eventually, use the index in the second level array to access each element in it.

By the way, use Data::Dumper to peek into the structure, then no data structure is complex any more. To use it:

use Data::Dumper;
...
print Dumper(\%hash);
[download]

print Dumper(%hash);
[download]