There is an interesting post over on GotSpeech.NET (VXML vs. Workflow for Speech Server 2007) that compares speech devolvement using VoiceXML vs. the Windows Workflow model available in Speech Server 2007.
Given that most of my work building applications in C# and ASP.NET for the Nuance Voice Platform (NVP) I’ve got quite a lot more experience with VXML than Workflow (or SALT). I partially disagree with him when he sites a "longer development cycle" with VXML. It is all about familiarity with the language and platform. But for the most part I think he makes good points.
I think the Workflow model is interesting but I’m wary of tying myself to a single voice platform. I much prefer the flexibility of moving between Nuance, Microsoft, Voxeo, etc. as needed. Each platform brings a different strength to the table and it seems like a bad idea to limit my options at this point.