What are the best practices for benchmarking different Java collections?

I have couple interesting Java collections in my hand such as:

1. http://code.google.com/p/guava-libraries/
2. Java 7
3. Java 7 concurrent collections
4. Scala collections
5. Homegrown collections that we have at some company

I wonder what would be the best practices to test these API's, from a performance and scalability perspective, i.e. which one is fastest, most scalable, performant, etc. Should I set with million(s) of random elements and use timer or something else? Just wanted to satisfy my curiosity and see which one would win.