Wow, I like that -- thanks much!

1. Guaranteed O(n) -- right now, the comparisons are about 120% of n with normative mergesort and 102% of n with quicksort.
2. I can pop those into the new array meaning the memory use will remain flat (rather than temporarily doubling).

my @bucket = ();
while($#Notes >= 0) {
    my $n = shift @Notes;
    push @{$bucket[$n->{terms}-1]},$n;
}
#Need 1D array back:
while($#bucket >= 0) {
    my $b = pop @bucket;
    push @Notes, shift @$b while ($#{$b} >= 0);
}
[download]

Works perfect.