> and using a different combo of bcf/bsf for each channel. It doesn't
> seem very efficient though

I use macros, which at least makes the text a little easier to work with,
although I'm not cycling through them 0-7. If I was, my code would look
something like

adch0  macro
         bcf    adcon0,chs0
         bcf    adcon0,chs1
         bcf    adcon0,chs2
         endm

adch1  macro
         bsf    adcon0,chs0
         bcf    adcon0,chs1
         bcf    adcon0,chs2
         endm

adch2  macro
         bcf    adcon0,chs0
         bsf    adcon0,chs1
         bcf    adcon0,chs2
         endm

adch3  macro
         bsf    adcon0,chs0
         bsf    adcon0,chs1
         bcf    adcon0,chs2
         endm

adch4  macro
         bcf    adcon0,chs0
         bcf    adcon0,chs1
         bsf    adcon0,chs2
         endm

adch5  macro
         bsf    adcon0,chs0
         bcf    adcon0,chs1
         bsf    adcon0,chs2
         endm

adch6  macro
         bcf    adcon0,chs0
         bsf    adcon0,chs1
         bsf    adcon0,chs2
         endm

adch7  macro
         bsf    adcon0,chs0
         bsf    adcon0,chs1
         bsf    adcon0,chs2
         endm

         adch0
         acquisition delay
         bsf     adcon0,go_done ;start conversion
         nop
         nop
         call    getdata
         call    do.something.with.data

         adch1
         acquisition delay
         bsf     adcon0,go_done ;start conversion
         nop
         nop
         call    getdata
         call    do_something_with_data

etc

;--------------------------

getdata

         btfsc   adcon0,go_done
         goto    $-1
         movf    adresh,w
         movwf   hi
         bank1
         movf    adresl,w
         bank0
         movwf   lo
         return

repeated for adch2 to adch7, using fsr to store the conversion data in
RAM. If you used a bit manipulation routine to cycle through the channel
select bits I'm not sure you'd gain much. It would / might take less memory
but would possibly take longer. Depends what you call "inefficient" - time
or code space ? You could integrate the "do_something_with_data"
routine into the acquisition or conversion time, which would make the
cycling "more efficient"

--
http://www.piclist.com hint: The PICList is archived three different
ways.  See http://www.piclist.com/#archives for details.