First, the constructive part. Crick, who is as physicalist on neuroscience as one can wish for, in Astonishing Hypothesis discusses "the processing postulate":
"It suggests that we may be using the words conscious and unconscious for two many somewhat different activities. They may have to be replaced by some phrase like "processing unit", or, in some cases "awareness unit". Each of these units will have its own semiglobal representation, usually covering several cortical areas... Each particular thalamic region may handle its own form of attention, possibly by allowing neurons in its member set of cortical areas to talk to neurons in the thalamus, which in turn feed back to them, so that in some way their firing is coordinated... This does not mean that thalamus by itself can produce all different forms of awareness. Awareness requires activity of the various cortical areas as well as the thalamus, just as a conductor needs the orchestra to produce the music".
The thalamus may or may not be hosting mind's conductor, but the idea is perfectly physicalist, and even compatible with a mild form of bundle theory.
Second, just because a phenomenon is not fundamental (and that is relative anyway) does not mean that it is an illusion as colloquially understood. There is nothing more to elastic waves than vibrations of atoms in a solid, but they are no illusions. Classical objects are specialized states of quantum flux produced by decoherence and persisting under certain circumstances, but they are no less real than the flux. Wallace in Everett and Structure calls this "the fallacy of exactness", and illustrates it with example of a tiger:
"tigers are unquestionably real in any reasonable sense of the word, but they are certainly not part of the basic ontology of any physical theory. A tiger, instead, is to be understood as a pattern or structure in the physical state... A tiger is any pattern which behaves as a tiger... There is a concept of transtemporal identity for patterns, but again it is only approximate".
Even optical illusions like rainbows are as real as light and water drops that produce them. We treat them as "illusions" for pragmatic reasons, doing otherwise would lead to inefficient behavior, but there is nothing "metaphysically illusory" about them. As Berkeley pointed out there are no illusions in epistemology. There is also much more to collective states (whether or not 'I' is one) than combination of disparate processes. That would be intense interactions between distant parts due to feedback loops for example, a glue that produces highly correlated behavior of the parts with or without the conductor. And whether it is a conductor or a glue that does the deed is in the end secondary.
To paraphrase Hume, there is no valid path from function to implementation. File sharing could be implemented with a central server, as Napster did, or without it, as did Morpheus and Grokster. Inferring restrictions on implementation from functionality of mind (e.g. "soul is simple and immaterial") is an illegal move that already Kant criticized rational psychologists for. One does not have to reflect the other, it can just emulate it. We did not conclude that mind's host is more like a neural network than like Turing machine from what mind does, the two are functionally equivalent, neuroscience employed empirical means for that. We can not conclude that mind is material, ideal or dualistic from what it does either, so materialism imposes no restrictions on functional models of self. Kant's model is perfectly compatible with it (as long as "materialism" is understood in the usual sense, not in Kant's own special sense), for all he says we know things-in-themselves may well be material.
