The EJB spec mandates that the access to EJBs be
thread safe. This means that multiple clients cannot be accessing the same instance of the stateless bean at the same time. One way to achieve this goal is to have a single instance of the bean and "wait" till each of the client completes the access to the bean before letting the next client in. This obviously isn't practical for real applications. That's where the pool comes into picture. A certain number of bean instances will be available in the pool and whenever a client wants access to the bean, an instance can be allocated from that pool. This way multiple clients will work on different instances of the stateless bean at the same time.
