@Dramanaught
Thanks a lot! No overselling has always been our workhorse, we were the first to publish a real time servers monitor as an element showing no overselling. Load balancing inside single VPN servers came later in 2017 (general load balancing at infrastructure level was built in 2011 and improved in 2012 and 2013).
About the load balancing at server level: since OpenVPN runs in a single core and does not scale, we decided to run one or two OpenVPN daemon(s) per each core, and send each new client connection to the OpenVPN daemon running in the least loaded core (according of course to the picked connection mode). As far as we know no other VPN service (based on OpenVPN) does the same. The outcome has been interesting because we can now squeeze up to 1.7 Gbit/s on AES-NI machines with 4 cores Xeon CPUs, which should be a record for OpenVPN servers on equal terms (CPUs etc.).
Kind regards