Yeah sswift, that has logic for blitz but things work considerably differently in shaders. They dont have to, but they do if you want speed and a good FPS.
Good FPS is relative, mind - if you can afford to render twice then you can afford to render twice, and with rather plain geometry being detailed by, say, dot3 these days (outside of B3D, of course) I'm guessing it's more likely to be affordable than ever before. That Gamasutra article seems to cover both methodologies, though, so I guess this is really just a case of settling into the process you're most comfortable with. If these things are available then you generally find yourself tinkering with them anyway - shaders don't really need evangelising I'd guess, purely because programmers like to explore new tech.
Node based shader editor. That means you make shaders for nodes (in blitz called entities) but they also include an option to see postprocess shaders on any object. You refuse to understand that postprocessing cannot be done on a single object. If you had a single sword rendered and scenery hidden, then the shader wouldnt be able to spread it's bloom over the background (like that wall up there) because the wall isn't rendered, now is it? ;)
It is not true that this isn't possible, and it's kinda what I was expressing back with my S\/\/ANKY water teasers when I said that it introduced masses of scene management. You can do all this stuff, but it costs (which at least validates some of what you've said about performance): Sticking with your example, one would render any scenery that obfuscates the object in question in such a way as to create a mask, then once you re-render normally and overlay the bloom everything would look dandy. Rendering a single object is a really bad way to think about the process - what you are actually doing is "focusing on" a particular object (or, indeed, collection of objects) and that means taking into consideration anything else that needs to be involved in the FX render.
EDIT: I think the swear filter went a bit mental!