I think there might be some gotchas with some of the Erlang clustering primitives. I worked on a product that Erlang solutions helped design and they didn't use Erlang clustering for RPC but rather reimplemented it themselves over dedicated TCP connections. If you do some googling there are people complaining about issues in Erlang RPC caused by multiplexing the RPC messages between nodes and cluster meta messages over the same TCP connection creating head of line blocking issues. Though, maybe some of this has been fixed since:
https://www.erlang.org/blog/otp-22-highlights/#fragmented-di...
There are also apparently other bottlenecks like funnelling everything through a single gen_server. (https://erlang.org/pipermail/erlang-questions/2016-February/...). Again, this may have been fixed since then.