并发性能提升的原因之一是使用了 ConcurrentHashMap，原来的 Collections.synchronized(new HashMap()) 存在同步锁的性能瓶颈；
如果你是用JDK 5.0或以上的版本，就可以直接得到这个性能的提升的好处了。
或者下载 Emory concurrent utilities backport 加到 CLASSPATH 中
http://dcl.mathcs.emory.edu/util/backport-util-concurrent/
当然还要努力，越快越好啦。。。
引用
The upcoming Spring Framework version 2.5 will bring a 200% improvement for concurrent access over Spring 2.0.*. I used Crazy Bob's Semi Useless Benchmark as a starting point. I fiddled with the code slightly to change the default behavior from "prototype" to "singleton", and I saw that spring 2.5 was faster than Guice for singletons.
Here are some benchmarks (there is variability based on the runs)
* Spring 2.0 vs. guice 1.0:
o Prototype: Single thread: Spring is 50-150X slower. Concurrent: Spring is 50-150X slower.
o Singleton: Single thread: Spring 3X slower - 2X faster. Concurrent: Spring is 5-10X slower
* Spring 2.0.6 vs. guice 1.0:
o Prototype: Single thread: Spring is ~10X slower. Concurrent: Spring is 5X slower.
o Singleton: Single thread: Spring 2X slower - 2X faster. Concurrent: Spring:Guice ~2:3
* Spring 2.14m vs. guice 1.0:
o Prototype: Single thread: Spring is 5-7X slower. Concurrent: Spring is 3-4X slower.
o Singleton: Single thread: neck and neck. Concurrent: Spring is 1-1.5X faster
* guice 1.0:
o Prototype: Single thread vs. Concurrent: Concurrent might be a bit slower than single threaded
o Singleton: Single thread vs. Concurrent: Concurrent is quite a bit faster than single threaded
详细请见这里
http://www.jroller.com/Solomon/entry/spring_2_5_perfomance_improvements