Yes, the barrel shifter does save a bit of code space, since there is no need for an external loop. Of course, there exists processors that can do n-step shifts/rotates without a barrel shifter too - then the instruction will take a varying number of machine cycles depending on step count.

And as noted, 2-bit fields doesn't allow any huge gains from the bit banding feature. But some gains could be had from doing two writes instead of having load + and + or + store. However, I would think the ARM chip will fast enough anyway. And testing a specific value of the two-bit fields could be done with two integer loads with check for zero/non-zero instead of having to mask out the value.