I looked a little bit into it.
@summisimeon replied to the post of contest #7 a little bit after some other replies and the reply was upvoted by 5 users from busy.org, an app he is also using. So I expected a high "community" part in the bounty hunt. Was wrong.

@summisimeon has earned 0.455 STEEM. 0.447 STEEM from the creator of the bounty and 0.008 STEEM from the community!

Next one is @minimining. Replied very fast, first comment, and got the first upvote with 20%. Still same distribution:

@minimining has earned 0.456 STEEM. 0.445 STEEM from the creator of the bounty and 0.011 STEEM from the community!

Both cases doesn't seem to have the same cause, it is weird nonetheless.

Something else:
Every vote is "taxed" by 1.2 SP.
https://steemit.com/utopian-io/@crokkon/how-will-the-hf20-dust-vote-changes-affect-the-vote-values
Upvoting with 20% is nice and kind but it is reduced by around 8%.
Thanks to Hf20 this distribution system is kinda expensive.

Also, I didn't know that, eSteem is taking 10% from the rewards as beneficiarie. I don't know how much worth the exposure to a different audience is.

A small suggestion: Giving us a chance to win SBI in that contest is already good enough, paying every entry with steem bounty is basically overkill, even with that good intention.
How about, instead of getting a bounty hunt up, sponsoring a random entry with one share?
Or, a little bit more complicated:
Every entry to this contest is also an ticket to a sbi raffle. Those will carry over to next lottery but once won that random sbi raffle all tickets are set to zero again.

This would put out one sbi at least per lottery instead of a bounty and it would give everybody an incentive to return to the contest.

Just a little bit brainstorming around here. Pretty sure there is still somewhere a flaw in it. But at least it reduces the dependency to a third party.