> This is what I've discovered... It takes 12 clocks/words

You can make it a table-read, and fill the gaps with other sub functions.
Under the presumption that you can make use of any area of 3+ words in size,
this method uses 6 cycles (incl call/ret) and 11 words.

        addwf   pc,f
        (1 wasted word)
        retlw   0
        retlw   1
        (1 wasted word)
        retlw   2
        (3 spare words)
        retlw   3
        (7 spare words)
        retlw   4
        (15 spare words)
        retlw   5
        ...

(I'm a bit rusty in PIC asm, so I hope that I got it all right)