>      - PIC's one cycle 120ns instruction is pretty fast, especially when
>      compared with 8051.  (however, aren't their accelerated 8051
>      look-a-likes that *are* as fast as PIC?)

Dallas makes a part that runs a clock of 33MHz (I think) and runs "90%" of
instructions in 4 clocks.  Unfortunately, that 90% is only a measure of
opcode space mix, not dynamic or even static instruction mix.  To understand
this, consider that the CPU needs a lot more opcodes for instuctions like

        mov a,r0
        mov a,r1

etc. (one-byte instructions--4 cycles execution) than for

        mov a,#imm
        mov a,direct

(two-byte instructions--8 cycles execution) or

        mov direct,#imm
        mov direct,direct

(three-byte instructions--12 cycles execution).

To be sure, many instructions would require two instructions on the PIC to
replace them, but PIC's handy skip instructions often more than make up for
that; further, the PIC's addition and rotate instructions can operate with
memory as a destination whereas the 8051-clones' cannot.