There's no design pattern available to ensure this QoS requirement. You must define a intensive analysis onto the designed system and measure the response time using tools or frameworks.
The response time for a given process or operation is associated with a number of factors that can increase or decrease that number, like the number of transactions, the concurrent users at certain time, the number of nodes distributable for the application (application server, database server, redundante servers)
After you mitigate what points of your design are decreasing the performance, you can do something with patterns, for example, using the Flyweight pattern to ensure pooling at your beans and pojos, or using Caching Service Locators to cache the remote references of yours Enterprise Beans.
But the exacly answer about how to mitigate those Qos requirements, envolves stress tests and very refined analysis.