All of this confirms why forwarding should require approval of the forwarding system's system manager, a feature which SM and possibly all other products lack.
I did not follow your notation completely, so I will rewrite it as:
user1@domain1 sends to user2@domain2 forwards to user3@domain3
but user3@domain3 is invalid.
Your server is harmed multiple ways by this mistake. Domain3 perceives your message as a directory harvesting attack. If the volume is non-trivial, domain3 may also perceive you as a DoS attack. If the bounce is delivered, User1@Domain cannot possibly make sense of a message that says "Your message to user3@domain3 cannot be delivered", because he never sent a message to users3@domain3. But you said that the bounce is not delivered, possibly because SM does not handle this unusual situation perfectly, so now Domain1 has raised your risk score.
Now imagine that User3@Domain3 does exist, but it is incorrect because of a typo when the forwarding rule was created. User3@Domain3 complains to his admin and causes your server and domain to be blocked by his organization, and maybe your server and domain ends up on the SpamHaus block list. Now all of your domains are having delivery problems to multiple destinations, and you have no idea how you got blacklisted in the first place.
Or maybe User2 is forwarding to his personal account, where company secrets and privacy-regulated data become exposed to Google's email sniffing engine.
Or maybe User2 is malicious or compromised), so his forwarding rule is exfiltrating data to an external entity.
Assuming that you feel unable to disable forwarding completely, a fallback strategy would be to the new feature for redirecting bounce messages. Send bounce messages to your admin account instead of user1@domain1. Then you can investigate why messages are being sent to invalid addresses. Maybe its a one-time typo by an internal user. Maybe its an invalid forward like your example. Maybe its an infected account doing a blast attack. Whatever the cause, you find out, and you can manually notify anyone else that needs to know, after you have identified the root cause.