Currently when Magnolia is deployed in clustered environment, cache content is kept separate for each cluster node as it is stored outside of JCR. This however means that cache updates, or more precisely forcing pages to be removed from the cache explicitly (because for some reason we know they were updated, like when adding page comment) will not work properly as such event will not be propagated to instance of the cache running on other cluster node.
The existing implementation will still work fine for general use case when we need to flush cache upon activation which is observation based and therefore propagated for cluster nodes.
For more other use cases, like commenting, we should provide an API that will do the same (propagate the flush request using observation) for flushing individual content.