To provide some additional insight on this topic, let us revisit the security foundations for Chia and Subspace.
Loosely speaking, the foundation for Chia can be stated as: Given massive honest storage space \Omega, it is hard for the adversary to “match” the space \Omega (so that the adversary can do double-spending) in terms of the space-time tradeoff even with near-optimal strategies such as Hellman’s attacks.
This foundation applies to Subspace as discussed in our research paper. But more importantly, Subspace enjoys an additional security foundation: Given massive honest storage space \Omega, it is hard for the adversary to “match” the space \Omega (so that the adversary can do double-spending) in terms of the space-time-IO tradeoff under various strategies. In other words, the required IO for many attacks is too demanding for large \Omega.
Next, let us use on-demand plotting as a particular example of attacking strategies. As explained by Dariia, the IO of k22 is about four times the IO of k20. So, we can mask 4 pieces with k22. which does not give the adversary conducting on-demand plotting any advantage in terms of required IO. In other words, k22 with masking 4 pieces is as good as k20 with masking 1 piece for on-demand plotting.