This does not mean separate processes, it simply spawns one "VM" per thread. Lobster VMs are pretty light weight (just a self contained object that has everything for a program to run). The advantage is that everything inside one VM can be single threaded, down to memory allocator. The threads then communicate by sending each other lobster values, similar to a "tuple space". I personally think this explicit communication style of concurrency is safer and easier to use than shared memory concurrency.

As for the lifetime analysis, see some sibling comments: I need to do a good write-up of the algorithm. I wasn't anticipating this HN post :)