Given that the private key range is known for each puzzle, I am not sure I understand the advantage that someone will have once the public key is known.
For Pollard's Kangaroo, you need to know the public key that you're trying to match.
OP was theorising that once they publish their transaction, someone could use Pollard's Kangaroo to trivially solve the private key in a matter of minutes and then publish their own transaction stealing their prize.
I'm not overly familiar with the performance of this particular algorithm or the available scripts for it... but if the actual winner just disables RBF and sends with a "decent" fee, the odds of their prize being "stolen" would be pretty minimal, I would think.
It was
reported that one Tesla V100 can check 715 M keys per second by using bitcrack. Assuming you can get google to rent you 176k V100's, I calculate a 1 in 488 chance that you will find the private key within 5 minutes. It was reported on that same post that a V100 can make 1430 Million "kangaroo jumps" per second (about 2x as many private keys tas than it can check using bitcrack). Assuming that the scope of what needs to be searched is the same, this would give someone a 1 in 244 chance of finding the private key within 5 minutes.
I am not sure how many V100 google has on its platform but 176k is a lot, but I calculate that many V100s as having a retail price of about $1.1 billion. If you spend more than 5 minutes trying to crack the private key, you will be spending more money than the value of the coin in the address.
You have to know how kangaroo program works. A single V100 can solve a 64 bit key using kangaroo in mere seconds. One cannot use kangaroo to crack #64 now because the pub key is not known. But once someone broadcasts to transfer the BTC from #64's address, the pub key will be exposed and someone can use a single GPU to solve for the private key in seconds. That is what the OP is saying. As others have stated, using the RBF with decent fee, will help from others "stealing" #64s key.