Also voted for complete removal. Thing is, trying to implement "device independent" or "social" DRM will land you in exactly the same set of problems we have with current crop of DRM implementations.
You will still need to attempt to technically prevent removal of this DRM. You cannot simply inject a watermark file into epub container and be done with it, while the actual content is still intact as plain text. The only way to prevent removal of restrictions is through cryptography. Which, as we all should know by now, does not work when your intended recipient and attacker are the same person.
The only way to technically prevent DRM removal (for a period of time), is algorithm obfuscation. The more universal the DRM standard (industry-wide), the shorter that period is going to be. Open obfuscation is contradiction in terms. Thus, to hold up the pretense of it actually working, even "social" DRM will need to be proprietary and it can never be industry wide standard (for a "couple of years plus" sort of period of time).
All in all, any sort of cryptography-based DRM creates all sorts of technical obstacles while solving not a single problem in constructive way.
I voted for "only if the DRM was removed completely?" but a better statement of my position would be "only if the DRM could be removed completely?".
I dont buy DRMed books where I can't remove the DRM.