A couple of thoughts:

- When timing queries, there are so many variables - one of the most important of which is I/O. Run any statement twice and it will run faster the second time because data was read from disk the first time and put in cache. Therefore, the second run did not have to read that data from disk and ran much faster. Any time you are using actualy time to compare the efficiency of statements, you need to run *each* version enough times to eliminate caching as a variable. You must run each several times because, since they probably have different execution plans, they will be caching different data.

- Setting something to parallel is no silver bullet. A lot of variables are involved here as well. What are your configuration options? How many threads are attacking the problem? What is your I/O spread? How much did the plan change? Is this a statement that *should* be parallel? A good analogy I've heard is home repair: If you try to use 40 contractors to put in a bathroom sink, it's going to take longer than if you have just one. You don't actually want the 40 contractors until you're building an addition.

- Chris