It's not just scoops over 4000, there is no definitive cut off, but definitely as the scoop number drops the effectiveness also drops significantly. But the cut off point for where it is no longer effective will drop to lower and lower scoop numbers as GPUs get faster.
The problem is that not all of the rounds require the same amount of work and since you only need the result of 1 scoop for each round, you only need to do the work until that scoop and then can exit the loop early rather than processing all 4096 of them (It starts at the end and works backwards). If the scoop for the round is 4096 you only need to process 1. If the scoop is 4000 you only need to process 97. If the scoop is 1 you have to do the work for all 4096 and currently processing power is not fast enough to compete with poc mining for those lower scoops, but at some point it could be.
I wouldn't call it a hack or an exploit necessarily, really its just a more efficient miner, and it doesn't hurt the network security, but it can have an impact on distribution that may be considered unfair because not everyone has access to this more efficient way of mining burst. I mean efficient in terms of nonces/sec not energy.
This is a single thread using a CPU. Just imagine the results if this were in parallel using GPUs. I don't know how to use OpenCL so this is as good as I can do with this but Im sure someone will be able to do this in OpenCL and take advantage. It may already be happening and could explain the huge sudden jumps in net diff followed by huge drops. My theory is they are switching their GPUs to burst during high scoop rounds and then switching their GPUs back to ETH or whatever when Burst scoops are low.