> The atomic reference counts are indeed more expensive, however with compile > time optimizations and and lent/sink can help removing unnecessary inc/dec.
Most of these optimizations are unsound in a multi-threaded setting. But even if it all worked out, there is nothing "convenient" about shared mutable state with concurrent access, it just doesn't work. Decades of heavy C# and Java use have demonstrated that. Sending isolated subgraphs around is the better idea and should be our focus.
