callCC oddness

for some reason

fac n = (`runCont` id) $
          callCC $ \k -> do
            when (n == 1) $ k 1
            return $ n * fac (n - 1)

is faster and uses less memory (for n=10000) than

fac' = product . enumFromTo 1

or [code]fac'' n = foldr1 (*) [1..n]

what gives /prog/?

>>12
According to the Intel 8086 Opcode reference, an add reg16, mem is 3 cycles, whilst an inc reg16 is 2.