Since they don't repeat, each subsequent number requires one less bit to store.

If we store each number using the minimal number of bites, we can pack the information into 63 bytes, which is 70% of using a byte per number. (30% savings)

$ perl -MPOSIX=ceil -e'
   my $acc = 0;
   $acc += ceil(log($_)/log(2)) for 2..90;
   CORE::say $acc;
'
63
[download]

There's still a lot of loss in there. Each number introduces a little bit of loss. By using a variable-base number, we only loss for the whole line.

If you use a variable-base number, we can pack the information into 58 bytes, which is 64% of using a byte per number. (36% savings)

$ perl -MPOSIX=ceil -MMath::BigInt -MMath::BigRat -e'
   my $acc = Math::BigRat->new(1);
   $acc *= $_ for 2..90;
   CORE::say Math::BigRat->new($acc)->blog(256)->bceil;
'
58
[download]