Probably half of IT projects fail. This study shows 80% of SOA projects failing.  But the article does give common threads among the initiatives that have succeeded.  KPI's experience is that C-level buy-in and a cultural shift are necessary for any SOA initiative to succeed (and that often comes with a change of CTO or a visionary one already in place).
Article on ZDNet about SOA increasing the cost of defects. McKendrick is completely wrong when he says the purpose of SOA is to "simplify things".  The purposes of SOA are integration, interoperability, and a competitive advantage in acquisitions.  SOA, of course, introduces new methodologies and new products into the organization.  This will in turn increase the number of errors.  Process is essential to SOA -- the earlier you find a defect, the cheaper it costs to fix.
Analysts and non-technical tech writers are using the term "cloud computing" with increasing frequency.   Cloud computing means too many things.  In most cases, analysts simply are clueless when throwing out the term.   Here is an article that attempts to define three types of "cloud computing".
At Network World, a panel conclusively agreed that SOA efforts are failing due to lack of process.  SOA is a new way of developing software and your process must reflect that.  Otherwise, your project's cost  will rise uncontrollably due to re-engineering efforts, something KPI has seen.
There is an article about Mule Galaxy Enterprise, the first open source SOA with integrated registry and repository.   KPI advises companies to stay away from any open source SOA solution.   Quite simply, SOA initiatives need good vendor support.
Dave Rosenberg is a master of the obvious when he states that Microsoft is missing the boat on SOA.   IBM comes closest to a single-vendor solution (WSRR, DataPower, WebSphere, Process Server, MQ) while Oracle is closing in with its many acquisitions (BEA, Tangosol).  For new SOA initiatives, KPI still recommends IBM.