Are you talking about the forgery problem? C/R at most doubles the volume of mail currently being handled. However, if your spam filter is already classifying 80% of the mail you get, which it had better, then C/R is only adding 20% to the total volume of mail (with most of the challenges discarded.)

I fail to see anything that doesn't scale about that. The only thing that doesn't scale is the autoresponses to forged addresses, if those victim addresses don't have their own suitable filtering. Beyond that, how can anything that simply does a small linear multiple not scale? "Not scaling" refers to something that gets harder to handle (per unit) the more units you have, not something that remains linear.