I had to read up a bit on (un-)bounded non-determinism there. It would seem actors give you even looser safety properties in that it has unbounded non-determinism and you can't really sure about anything about your programming model in the default case.
That was the point, you can't have something reliable and performant in a real world if you use a bounded model for it [world].
I'm actually don't agree at all to this. Is e.g. the MillWheel software, or Kafka, or Storm/Spark inherently unreliable and non-performant because they feature strong consistency and bounds on their operations? I don't think so.
So, basically fighting actors with channels and synchronous selective operators, right? This is not how it works, because bounded nondeterminism is precisely what actor model was trying to get away from.