Why reinvent the wheel, my suggestion will be to use existing framework for your caching mechanism and leave all set of caching problems for framework.
I will like you to point to
Java Caching System
This is a subproject of Turbine and is available on jakarta site.
some of the high points are listed below.
JCS is a distributed caching system written in java for server-side java applications. It is intended to speed up dynamic web applications by providing a means to manage cached data of various dynamic natures. Like any caching system, the JCS is most useful for high read, low put applications. Dynamic content and reporting systems can benefit most. However, any site that repeatedly constructs pages, dropdowns, or common search results form a database that is updated at intervals (rather than across categories continuously) can improve performance and scalability by implementing caching. Latency times drop sharply and bottlenecks move away from the database in an effectively cached system.
The JCS goes beyond simply caching objects in memory. It provides several important features, necessary for any Enterprise level caching system:
Disk overflow (and defragmentation)
Quick nested categorical removal
Fully configurable runtime parameters
Remote store recovery
Non-blocking "zombie" (balking fa ade) pattern
Optional lateral distribution of elements via HTTP, TCP, or UDP
Remote server clustering and failover (almost complete)
These features provide a framework with no point of failure, allowing for full session failover including session data across up to 256 servers.