However, what I do not quite understand is, what's the difference between the upcoming PubSub mechanism and the current "subscribe" mechanism.
If, for example 10 connections connect to the same OPC server and subscribe to changes of an item, how is that different from the new PubSub mechanism?
UA PubSub in its current draft form defines how to plug an external message queue into an OPC UA server.

For this, DataSets are defined, who define a mapping of attributes in the UA information model onto Pub/Sub messages. MessageWriters and MessageReaders (de)serialize the messages coming over the external message queue and sync the content with the UA information model.

If you just want to read 10 variables with pub/sub, the existing subscription mechanism should work just as well.