In the context of deciding if BizTalk Server 2004 is included in a workflow software development solution:
Traditional software architecture and design approaches try to get the strategic decisions right up-front, as all-or-nothing design decisions (e.g. BizTalk—yes or not—for the entire project), but design of software has not necessarily to be that way, or not the only way. Yes, it is completely understood that the context usually is a business context, not an engineering context or a context of applied science; but, reflecting on how the bottom line _sometimes_ ends up, the better business case comes from a co-evolution model of problem-solution design with on-the-fly adaptation accordingly to confirmed current (and changing) business needs. Of course, this needs a proper context where chances of success are present.
The BizTalk Server 2004 high-level design is sound; it addresses the scalability and performance problem very nicely. See if your target design goals match with those of BizTalk, if that is the case and you still see BizTalk HWS as a not a must, consider plain BizTalk orchestrations perhaps end up making a lot of sense, perhaps not –few things are for sure when you are short of experimental, in-context, data.