From Bryan Dollery on the Extreme Programming mailing list:

If business leaders were interested in money then they'd use JBoss overWebSphere or WebLogic. The three products are, for most uses, identical -but JBoss is free while Web* can cost around $50,000 per processor. If money is that important, why do these products sell?

I make a living as a Websphere consultant, so I'm going to put my devil's advocate hat on. Please note that I really do like JBoss, I just don't get paid to use it. I don't want to start a ‘My Appserver is better than yours!’ battle, I just want to demonstrate that there are reasons that Websphere and its ilk exist in the marketplace, even though its free (or at least orders of magnitude cheaper) competitors are generally more up-to-date with the standards, easier to use and faster.

I can't speak for Weblogic, having never used it in anger, but as far as Websphere goes:

• Trust is an asset (1). In the J2EE Container Shootout, JBoss's Marc Fleury said that of his competitors he'd either choose Orion because it's superior technically, or Websphere “because IBM will be around for ever and ever - it is always a safe choice.” As good as Open Source support generaly is, a support contract is a far more tangible asset.
• Trust is an asset (2). JBoss is around the same place Linux was five years ago. It's the same Catch-22. To be trusted, it has to be seen running critical applications, but to be deployed in critical applications, it needs to be trusted. Linux managed that through two prongs, firstly by being deployed in thousands of Internet providers who didn't want to pay for commercial Unices any more, and secondly by system administrators sneaking Linux boxes in while nobody was looking, so a year later they could say “Yeah, we run Linux, it's been delivering your mail the last twelve months without a hitch.”
• Scaleability. Clustering is a very new feature in JBoss, and it's hideously under-documented. I'm told it's very good, but all I can get from the JBoss site is “We have clustering, but to find out anything about it you have to buy our book”, which doesn't fill me with confidence.
• Documentation. Giving away the product and then hoarding the documentation is just plain stupid, not to mention being very anti-GNU. Remember, software is only free if your time has no value, and the time of most IT contractors is very valuable indeed. Hiding information on how to use your product adds a high hidden cost to its use. Contrast the JBoss clustering book with the Workload Management Redbook for Websphere 3.5, free for download, and 600 pages long.
• Bundling. This is the biggest reason. Most people who buy Websphere don't buy it in a vacuum. They're buying a package deal of hardware, software and consultancy. The cost of the individual Websphere licenses generally become part of a big lump-sum that covers the entire project from conception to post-deployment support. That big sum gets negotiated, and divided up internally, but to the customer it's only really the big number that matters. And bluntly, if it weren't for the software margins, the programmers would be far more expensive.