JPEG met the novelty, non-obviousness, and usefulness tests, as did LZW. That they're considered nearly trivial these days is because they were such influential algorithms that they've become part of the programming canon.

That said, please don't represent my stance as a matter of a cutoff. Many things in this world, including the question of patentability, are fuzzy distinctions that require some subjective "good judgement." Common law legal systems are bad at handling this, which is partly why judicial philosophy is so complex.

If I oppose patenting an algorithm/data structure, it's because I don't think patent protection is necessary in order for people to push the state of the art forward.