Performance considerations

While PHPCR can perform reasonably well, you should be careful. You are working with an object model mapping interlinked data. Implementations are supposed to lazy load data only when necessary. But you should take care to only request what you actually need.

The implementations will also use some sort of storage backend (Jackrabbit, (no)SQL database, ...). There might be a huge performance impact in configuring that storage backend optimally. Look into your implementation documentation if there are recommendations how to optimize storage.

One thing not to worry about is requesting the same node with Session::getNode or Node::getNode/s several times. You always get the same object instance back without overhead.

Only request what you need

Remember that you can filter nodes on Node::getNodes if you only need a list of specific nodes or all nodes in some namespace.

The values of binary properties can potentially have a huge size and should only loaded when really needed. If you just need the size, you can get the property instance and do a $property->getSize() instead of filesize($node->getPropertyValue). Any decent implementation will not preload the binary stream when you access the property object.

When getting the properties from a node, you can use Node::getPropertiesValues(filter, false). This allows the implementation to avoid instantiating Property objects for the property values (and saves you coding). The second boolean parameter tells wheter to dereference reference properties. If you do not need the referenced objects, pass false and you will get the UUID or path strings instead of node objects.(If you need one of them, you can still get it with Session::getNodeByIdentifier. But then the implementation will certainly not be able to optimize if you get several referenced nodes.)

But request in one call as much as possible of what you need

If you need to get several nodes where you know the paths, use Session::getNodes with an array of those nodes to get all of them in one batch, saving round trip time to the storage backend.

You can also use Node::getNodes with a list of nodes rather than repeatedly calling Node::getNode.