It's not as much "non-exit" as Tor middle relays. Because it just connects to the VPN2 server using OpenVPN over standard TCP/IP. Instead of some proprietary protocol. But at least it's locked down with pf rules, so that it can only connect to the VPN2 server.
The diagram shows a nested chain with just two VPNs. But you can add more layers. As I recall, as many as six or so. Latency goes up, and MTU goes down. But throughput doesn't crash as much as you might think. I don't know why. But maybe it's caching.
So basically, you have a NAT chain locally in VirtualBox or whatever. And each NAT router includes a remote VPN server.
In order to share it, you'd need to open a port for incoming OpenVPN connections. Either locally, or forwarded to one or more VPN servers. And then you could route traffic through another VPN server in the chain.
