Yes, I agree that BEA's Cajun violates the write-once-run-anywhere ideology, but I also think that such vendor specific extensions are one of the reasons for Java's strength. Sun is in the process of releasing the specs for version 1.3 of J2EE, and it has said that a framework for web services will not make it's way into J2EE until version 1.4. In the meantime, BEA has decided to move ahead with it's own web services extensions in Cajun.
Java evolves when vendors like BEA spend R&D dollars on technology such as Cajun, and they submit such technology as proposals to Sun for future versions of Java and J2EE. If Sun were to carry the entire burden of advancing Java on it's own, Java would never have advanced as quickly as it has. For example, IBM was a major contributor to J2EE, and IBM has more Java programmers than Sun does. With this open model, Sun is leveraging economies of scale in spreading R&D costs similar to the ones Microsoft enjoys by virtue of the its massive installed base of Windows.
Something similar happened with relational databases. Most relational databases implement SQL, but vendors have added their own extensions and tools over time to gain differentiation. However, SQL does not have a guardian like Java has in Sun.
Any customer who chooses to go with non-standard technology must realize that there will come a time when s/he might have to abandon it. But for those early adopters who can?t wait, the early experimentation and learning might prove tobe a competitive advantage when the standards eventually arrive, especially if the standards incorporate some of that with which they were experimenting.