The first version of factorial in SBCL takes less time and processor cycle,

  CL-USER> (time (fact 5))
  Evaluation took:
    0.000 seconds of real time
    0.000002 seconds of total run time (0.000001 user, 0.000001 system)
    100.00% CPU
    1,550 processor cycles
    0 bytes consed

   120

compared to second version of factorial

  CL-USER> (time (fact2 5))
  Evaluation took:
    0.000 seconds of real time
    0.000027 seconds of total run time (0.000027 user, 0.000000 system)
    100.00% CPU
    64,240 processor cycles
    0 bytes consed

  120

How it works for other lisp implementation. However when I reach for bigger number things looks different.

  CL-USER> (time (fact 10000))
  Evaluation took:
    0.044 seconds of real time
    0.045335 seconds of total run time (0.036745 user, 0.008590 system)
    [ Run times consist of 0.021 seconds GC time, and 0.025 seconds non-GC time. ]
    102.27% CPU
    111,336,538 processor cycles
    69,739,552 bytes consed

  CL-USER> (time (fact2 10000))
  Evaluation took:
    0.021 seconds of real time
    0.021083 seconds of total run time (0.020312 user, 0.000771 system)
    [ Run times consist of 0.003 seconds GC time, and 0.019 seconds non-GC time. ]
    100.00% CPU
    51,983,719 processor cycles
    78,751,760 bytes consed

I wonder some kind of compiler magic kicks in, for higher number.