I'd imagine for small n and b newpgen and twinsieve will take roughly the same time. For larger values of these quantities twinsieve will likely have the advantage as 1. It doesn't have the memory restrictions that newpgen has, and 2. It's part of the mtsieve framework, so we can run it multithreaded.
And it appears your list for b = 20 is missing two primes: the ones for n = 8 (k value is 3) and 68 (k value is 60).
