SOA on ACID

Tony Baer reported a recent IBM event where Steve Mills talked about how the next stage for SOA was to add ACID reliability and fault tolerance to it.  Am I the only one that doesn't understand this?

If IBM really means ACID (as in formal database-type transactions), then maybe they need a bit of re-education.  It's commonly agreed that ACID is not appropriate for a loosely coupled environment.  Instead, guaranteed messaging and compensation are the appropriate means.

But, if Steve Mills didn't mean ACID - and just means "reliable", then I'm also confused because there are a lot of organizations doing this today.  We have dozens of customers running high volume mission critical applications - where revenue depends on the reliability of their transactions... and they are doing this on SOA.

Now, I suspect that IBM isn't really planning on trying to run every application inside DB2 to get ACID properties (I suspect only Larry Ellison would propose a vision like that).

So, maybe, and this is just a thought, maybe what they mean is that when you stitch together 60+ CDs worth of different products and call it a "unified solution" (after all, they all have the same brand, so it must be a unified solution), it might not be the most robust approach.  If this is the case, I applaud IBM's desire to reduce the complexity of their product lines... but let's not talk about this as a deficiency of SOA infrastructure.