Generally, we don't have to care about the Puppet Master's performance when we have a few nodes to manage.
Few is definitely a relative word; I would say any number lower than one hundred of nodes, which varies according to various factors such as the following:
System resources: The bare performances of the system, physical or virtual, where our Puppet Master is running are obviously a decisive point. Particularly needed is the CPU, which is devoured by the
puppet masterprocess when it compiles the catalogs for its clients and when it makes MD5 checksums of the files served via the fileserver. Memory can be a limit too while disk I/O should generally not be a bottleneck.Average number of resources for node: The more resources we manage in a node, the bigger becomes the catalog and the time required to compile it on the Puppet Master, deliver it via network, and finally receive and process the clients' reports.
Number of managed nodes: The more nodes we have in our infrastructure...