OpenRefine's current architecture only support standalone which means it has its limitation such as your server capacity, JVM, file size, data complexity and browser limitation etc. It is very hard to give you a straightforward answer for you questions due to those combination.
This is one reply regarding this:
There are some projects trying to scale the openrefine by clustering it:
There was some discussion to integrate with HDFS/Spark. But the effort is non-trivial to implement this. If everybody want to explore more on this option, let me know.