Your team has deployed a microservices application on Google Kubernetes Engine (GKE) with multiple services communicating via internal ClusterIP services. You notice that some requests between services are failing intermittently with 'connection refused' errors. The services are defined with readiness probes. What is the most likely cause?