Good thoughts. A few things:
1.) This would require a big media company like the HuffPo to purchase a lot of Steem Power. I don't think any big media publishers are willing to hold a legally ambiguous digital asset on their books anytime soon, though maybe a startup trying to disrupt would.
2.) A publisher that did this would still be subject to the whims of other whales (right now @steemit owns $66M, @ned owns $3.5M and @dan owns $2.7M worth of steem power). A large, premium publisher would probably rather fork their own blockchain than use a blockchain that gives up so much power to others (at least for now).
With regards to 1, definitely. I could see them making a multi-hundred thousand dollar investment in Steem Power instead of investing in as much community management and recruitment required when trying to get experts to contribute content for free, on a deadline. I hear you on the legally ambiguous nature of it and that being a risk.
And as for 2), I'm not sure why they'd be subject to the whims of other whales. On their own platform they'd curate whatever content they wanted displayed, so the whales voting would have no impact on their ability to show content they wanted. The rewards for the authors would be jumpstarted by the publisher's Steem power enabled votes, and if additional whales voted, great! But if not, then there's still upside from other platform voters. If the content is great then theoretically the Steem protocol and incentives should do their job and the writer will get adequately rewarded over time. If rewards are solely dependent on whales' votes, then Steem protocol has a bigger problem.
That's a good point on 2. I guess I was assuming that all Steem power is fungible (like all BTC is fungible).
If a publisher built a service on top of Steem that curated to their liking (and essentially mitigated the power of the Steem power of some user), I'm not sure how much success they would have. I think fungibility is pretty important to the success of any service on top of Steem, but could be wrong.