Our declarative pipelines use this:

    agent {
        node {
            label 'my-label'
            customWorkspace "/path/to/workspace/work"
        }
    }

where the customWorkspace points to a nfs share.

It correctly creates the @1, @2, ... directories as needed.

When the label only matches one node we never see a problem, but if that label matches many nodes we see it use the same workspace (with our without @n) for builds running at the same time on different nodes.

I tried to get around this using NODE_NAME in the customWorkspace but that is null as mentioned in JENKINS-66289