They used transaction hashes. Those are different, as they can be pre-computed by anyone before sending the transaction. There's no one else to blame but them, for using a mechanism they didn't understand in a critical part of their system.
Block hashes are way harder to manipulate, it would need to be done by the witness producing that block, which is not really trivial to do (calculate hash with the tx included, if win publish the block with, if not publish it without the tx) and as you correctly said should absolutely result in downvoting him.
If you use the hash of a later block than the one the transaction is included in there's no way to cheat it, and Dan's suggestion would've been totally valid.
//edit: well, I guess witness collusion would be theoretically possible, if the witness following the one including the bet publishes a block with a fitting hash. That won't always work though, as it requires quite some calculation to find a fitting block, and during that time you may miss the time window to publish. It would also require at least one of the two to be a top-20 witness)