BTM is recalculating the diff every 720 blocks, with an steady hashrate that translates to 1 day.
Imaging clever jumping in for exactly 1 interval of 720 blocks and than leave normal miners with the skyrocketed hashrate for the next interval for the following 720 blocks. Scary to me. It could take many days to solve those 720 high diff blocks.
That although opens the gates to malicious hashrate attacks.

DGW3 is adjusting every new block based on the last 24 blocks with a max increase to 3 times of the previous or max decrease to 1/3 of the previous. Reducing the interval of 24 blocks can make it adapt quicker (less smoother change), but on a jump you are still left with 1 block calculated based on the hashrate before the jump (called 'bad block' by me).
If we have a chance to see a bigger hashrate chance already during a block, we could take that into the calculation and the 'bad block' is gone. We should still set bounds of min and max block times and with some thinking we should even be able to translate that into a diff based on the actual hashrate.
Rejecting blocks sounds bad, resending blocks with a corrected diff sounds far better if possible. This could help at the beginning of a spike but sadly not at the end.