database connection pooling, friend or foe

So, I've been mulling over database connection pooling quite a bit recently. Our performance has been a bit sluggish, and I think how we're handling connections is a big part of it. On the one hand, pooling saves us from having to create a new connection every single time, which *should* make things quicker. But, I've also seen times where bad pool settings or certain kinds of jobs caused problems, like running out of connections or even deadlocks. How is everyone else handling their database connections? Are you guys using pooling? What are the main pitfalls you've hit, and what do you do to keep everything running well?