Yes, I am sure that you put exactly your finger on the issue.
Since the output is the same on the remote host than on the SSH slave sessions, the problem is not Net::OpenSSH or even Perl related. It is really about this busybox traceroute implementation on the host and therefore outside the scope of this very helpful place ;)
I was able to solve (avoid) the problem by having the Smokeping using only one instance at a time for that custom probe, so no multiplexed ICMP return packets can be mixed up by traceroute on the host.
By the way, this Net::OpenSSH module is great (thanks Salva for it) and use very little CPU on the local device and the remote host. I started by simply changing the TraceroutePing probe to use ‘ssh user@host traceroute etc...’ instead of ‘traceroute etc...’. It worked but the toll on local and remote CPUs was very bad.