A new wave of SOA "anti-hype" commentary is appearing in the media. Now that more and more companies have real-world experience, we're learning that SOA, no matter how great it may sound in theory, takes a lot of painstaking work to get to deployment.
At the top of the list is the problem of pitching SOA to upper management. It's hard to explain this thing that's not a technology, not a product, and ultimately delivers pretty vague benefits, like "agility."
Then, after you've sold SOA to management, you have to sell it to the developers. Not an easy prospect, because developers (understandably) like to have control over their work and not depend on the coding of others to get the job done.
The disjointed reality of enterprise data, with its multiple definitions for the same category (e.g. "customer") and multiple schema, is another problem. To date, most SOA teams seem to be treating data as an afterthought.
Finally, many commentators say vendors aren't helping. Specifically, vendor-based methodologies are typically more about the wonderfulness of the vendor's product than they are about solving the IT department's problem.