I suggest you test it with your code. It can make a different to the compiler in some cases however it is more likely to make a difference to the JIT. When it comes to micro-benchmarks and micro-tuning what should make a difference and what actually makes a difference is often not the same thing and only good testing is the way to be sure.
Another problem you have is that the JVM is improving all the time and a trick which made a big difference before may no longer apply. e.g. in Java 5.0 Lock was much faster than synchronized however in Java 6 the difference is much smaller and synchronized can be faster.
In generally, its a good idea to make your code simple, clear and easily maintainable and this will also result in efficient code.
