Thank you for pointing to iptables rules. This was likely the cause. I was unable to solve it fully, though. The downside of my solution is that I can just choose between a working internet connection for me (the spoofed machine can't effectively access the network) and working internet connection for sys-net the spoofed machine (all VMs except sys-net have networking unavailable).
I just did these steps:
1. Backup iptables rules using sudo iptables-save, piping the output to a file.
2. Remove all the rules (maybe just rules related to NAT would be enough):
http://www.adminsehow.com/2009/08/how-to-clear-all-iptables-rules/3. Perform all sslstrip steps (enabling IPv4 forwarding, arpspoof, one iptables rule and sslstrip itself).
4. Do some adjustments (one might run Wireshark, I did some tuning with Nginx)
5. After everything is done, run sudo iptables-restore < backup-file. Rebooting sys-net would likely do the same job, but I would have to reboot other VMs, too.
Unfortunately, even the inter-VM networking does not work in this case. I wanted to run Nginx in an extra VM connected directly to sys-net, but it was unreachable from sys-net. So, I installed nginx directly to sys-net. It is not optimal, but it is acceptable for me.
As network spoofing is not my daily job, such limitations (e.g. networking in sys-net only during the attack) are acceptable for me.
Regards,
Vít Šesták 'v6ak'