If all it takes is 100k GPUs (and I believe this to be true) then the answer to the time it takes to brute force is not 'Infinity' or 'almost impossible'.
There are millions of GPUs in the world. Think about this, the group here and at least one other group has created a pool in the past to crack one of the puzzles. Imagine if this group can find 100k GPU owners to join. Impossible? Not really, imagine a black hat hacker figures a scheme to create a virus that is highly viral. He could in theory infect 500k computers with GPUs and bruteforce the key. Highly unethetical and illegal, but that's not the point. The point is that there are ways to do it. So, can it be done? Fortunately, bruteforcing a key here and there is different from figuring a way to crack any key which would be the holy grail to killing bitcoin or at least forcing a change in the technology.
Another point I want to make - some of you have alluded to needing new technology, so something you may want to look at is the distribution of the random number generators used.
1. Imagine if the puzzle creator did use some random number generator, one could in theory figure out the distribution using the already cracked keys and target the main area of distribution.
2. Some of the tools used to crack the key uses random number generators, figure out the distribution and find a way to make it more linear or alter it to match the distribution of the rnd() used by the puzzle creator.
I just calcualted how much time does it need to crack puzzle 66 by bruteforcing
The answer is: Infinity OR almost Impossible.
What's insane is, even if someone managed to get 100,000 GPUs, you still would need 3.5 Months to go through all the 66 range. Not to mention electricity cost of the 100,000 GPUs. I think this puzzle is still infeasible. It's worth to mine better than trying to solve this. Or if Satoshi increases the prize again by 10x, so for puzzle 66, 166.6BTC, for puzzle 67, 166.7BTC and so on. THEN it will be worth investing huge mining army to destroy it. Other than that it makes less sense.
What's feasible though is puzzle 125, because of its public key.