Earlier quoted context omitted.
I believe that L2 would be weaker than L3 in practice, which likely explains why I've also never seen it implemented (If you know about an L2 instance I would be genuinely interested in taking a look)
> I believe that L2 would be weaker than L3 in practice How come?
In L3, you can obfuscate the cryptography and the video codecs together as a unit, blurring any defined border between them. A determined reverse-engineer can inevitably unravel that obfuscation, but it's non-trivial.
In L2, there must be some interface between the hardware and software components. That interface presents itself as a very obvious weak point. As an attacker, all you'd have to do is watch the data flowing out of the cryptography hardware, and into the video decoder software, and you'd be able to siphon out the plaintext video data.
(To be clear, this is entirely "in theory" because I've never seen an L2 implementation)