In strained situations, more resources than you asked for might get locked. This creates a bottleneck problem related to resources but it cannot be solved by adding more resources.
In worst cases will always lock all resources with a certain label instead of the required amount.
You can recreate the issue by configuring a couple (say 4) of lockable resources with a tag. Trigger a bunch of these jobs with NUM_RESOURCES between 1 - 4 to build up a queue of jobs waiting for their resources to become available. Eventually the error will start appearing.