First of all, I would use the v7 because of better cpu load distribution among cores.
For you heavy nat traffic, I would use the 1072 because of more cpu to process the nat entries.
You didnt show any figures about the current cpu load. Wich is the current installed Ros?
in ccr1072 average cpu usage It is the result of averaging the use of the 72 cores, which tends to deliver extremely low values even when you may have some cores with a lot of load.
in ccr2216 this average calculation is done only by 16 cores which leads us in most cases to a higher but more realistic value
so that numbers have a lot of room for interpretation
i think on the paper on pure CPU power ccr2216 is inferior to ccr1072
but
is easier to take advantage of ccr2216 cpu resources because they are not spread across 72 cores of only 1ghz with in-order-execution architecture
for single core tasks one ccr2216 core can do almost the same work as 3 cores of the ccr1072 because have twice the clock rate and is out-of-order execution architecture
I think ccr2216 has a Strong CPU which will be better for some scenarios like BGP Convergence times, but it can exist some scenarios where is not superior to a ccr1072
The only scenarios where ccr2216 is an indisputable champion is using Hardware L3 Offload, but it has some requisites and restrictions, also is a feature which is still improving
using Hardware L3 Offload you can have very low CPU usage even at Full Wire-speed
The very first step I would take, I would upgrade the router to the latest v7, so we can benefit of the missing route cache (less issues), the better kernel and a better load distribution over the cores.
You have a very old and vulnerable ROS version.
be aware all ccr1xxx will be discontinued this year so if you want to buy one do it now
porting your actual config with good results to ccr2xxx will not be automatic you will require to rebuild that config using single bridge vlan filtering and hardware offload, if you want 1to1 replacement better stay with ccr1072
maybe you dont need to update to v7 mandatorily, but at least you have to update to latest v6 to be more secure
table at the end. Doing a lot on the switch chip means - doing a lot LESS on the CPU to start with.
It also says on the table before (which handles the DX8525 switch) that the most active parts are offloaded, so you may end up in NAT with the heavy hitters.
So, I think it is not really fair to compare CPU when the main point of this machine is the ridiculous amount of hardware offloading - and a lot of that seems to benefit you directly.
If that is not working, I would consider getting a cheap I86 CPU, plug in 25g card or 100g card and run the routing in a virtual machine with like 16 or 24 high end cores.