If it turns out that these registers are loaded ONLY from the chip's data I/O pins D[7:0], then you don't need a mux; all you need to do is decode the control stuff to load the registers at the right time.

Driving the various outputs on the data I/O pins will require a mux and a tristate. The control logic for the mux should be kept separate from the tristate logic.It'll be something like this:

assign oe = (!RD_l && !CS_l);
assign dio = oe ? diomux : 8'bZZZZ_ZZZZ;

always @(*) begin : diomuxselect
    case ({A, ctrlword})
        CASE0 : diomux = ...;
        CASE1 : diomux = ...;
         ...
        CASEN : diomux = ...'
    endcase
end // diomuxselect

Note that diomux must be declared as an 8-bit reg.
Writing is similar.
-a