I understood this. What I mean is that if you store your private key exclusively on paper, storing it with 2-of-3 secret sharing still leaves a vulnerability when the shares are imported and combined (mitigated by combining in an offline clean computer, but still). If you have a 2-of-3 multisig address, and store the keys for that on 3 distinct pieces of paper, you get more security.
@Meni,
I think you're getting stuck on the idea that I have in some way proposed this idea as a replacement for multi-sig wallets. This is not intended to replace multi-sig. I'm not comparing it to multi-sig. It's nothing to do with multi-sig, at all. 0% related. It's simply an alternative for backing up your regular single-sig wallet. Your single-sig paper-backup has all the security risks you describe, but it's still an important use-case that many users (and maybe even organizations) will use even when multi-sig is available.
Multi-sig is completely unrelated to this, and is still my number one goal for the new wallets. Though, I guess you could technically create a 2-of-3 multi-sig wallet between multiple devices or parties, and each party could back up their own wallet using a M-of-N split-backup... sounds complicated. In fact, I might try to discourage fragmenting multi-sig-wallet backups because of it being complicated (but there's no reason it couldn't still be done)...
Use-case 1: simple user wants to use Bitcoin, and wants to have it backed up, but doesn't want any one user to have control of it. He may create a 3-of-6 split-backup of his single-sig wallet, keep one, and give the others to five friends who don't know each other (and don't know who they'd even contact to find the other pieces).
Use-case 2: These super-paranoid folks who think that someone at the bank will snoop in their safe-deposit box. Okay, so use two safe-deposit boxes at different banks, distributing 2-of-3 pieces (keep one, put the other two in two different banks). The likelihood of snooping is ridiculously low, but apparently still enough for the paranoids. However, that stupidly-low probability is squared when it's two unrelated banks so even most tin-foil hatters would be satisfied.
Use-case 3: Simply hide a few pieces around your house. As ben-abuya said, finding a single one won't be sufficient, but in 10 years from now when you need it, you'll surely be able to remember at least 3 of the 10 places you split it
(hell, hide one at your grandmother's house in one the 10,000 books she has on her bookshelves -- she doesn't even have to know or care that it's there)