Yeah my only point is that if we get rid of the bound on the number of conflicted transactions altogether, then we are effectively removing the bound on the number of elements in a diagram check (or limiting it to whatever the max number of inputs you could fit into a standard transaction multiplied by the cluster count limit).
So in this case, we might have to bring down the maximum number of transactions in a cluster as our only lever to cap the amount of work we might have to do…?