This isn't a completely new system: it's been tested conceptually already in BitShares. We plan to test it heavily before releasing it to look for any bugs that may be introduced in the port of the algorithm to Steem. "Live" testing shouldn't be relied on to find bugs.
Also, it's not like huge funds will be immediately at risk once it goes live, because the pool fills up gradually from inflation.
Someone else suggest an idea similar to the one you mentioned: create a version of a post that pays to the proposal funding account (instead of redirecting the burned/declined payout post). A more general version of that could be to actually allow users to create posts that payout to another user.
One advantage of using the declined payout posts versus either of the above options would be less code changes (including no real interface changes), but it also has the disadvantage of disallowing users from actually burning the rewards.
Sounds like a pretty good place to start to me. Plus this fund will be able to take donations so funding drives will be easy. I am pretty sure a lot of us would not mind donating an up vote or two per week especially as people see it working to better Steem's eco-system.
People were already donating votes to the burn post every day, why not just have them donate to a development post every day?
Because those quantities would never pay for any serious development.
I am against pulling 20% from the author reward pool. Pull some from every category and there is less of a dramatic effect on any one group.