We are duplicating lots of functionality in Seam that any JAX RS implementation has. We should probably drop SeamResourceServlet and maybe even SeamFilter in Seam3 and rebuild what we need on top of JAX-RS. The downside is that we'd create a dependency on a JAX-RS implementation. OTOH almost anyone might want one of these in the future anyway.
Yes, I think we should do this. JAX-RS has such a low footprint and is vital to the future of Java EE web apps, so like many other specs that we "recommend highly" I think this should certainly be one of them. We can always back off and try to abstract an SPI if the community is uneasy about the direction.