Definitely *not* a good benchmark. But less ridiculous than http://jabsoft.io/2013/08/29/why-c-still-matters-in-2013-a-simple-example/

fibo.c

#include <stdio.h>
#include <stdlib.h>
typedef unsigned long ulong;

int main(int argc, char *argv[])
{
    ulong n = atoi(argv[1]);
    ulong a = 1;
    ulong b = 0;
    ulong t;
    ulong i;
    ulong m = 1UL << 31;

    for(i = 0; i < n; i++) {
        t = b;
        b = (a+b) % m;
        a = t;
    }

    printf("%lu\n", b);
    return 0;
}

fibo.lua

local fib = function(n)
  local SZ = 2^32
  local a, b = 1, 0
  for i=1,n do
    b, a = (b+a) % SZ, b
  end
  return b
end

print(fib(tonumber(arg[1])))

fibo.py

def fib(n):
    SZ = 2**32
    a, b = 1, 0
    for i in xrange(n):
        b, a = (b+a) % SZ, b
    return b

import sys
print fib(int(sys.argv[1]))

results.txt

[fibo 15:45] n=100000000
[fibo 15:45] time ./fibo $n
1819143227

real  0m1.251s
user	0m1.244s
sys	0m0.003s
[fibo 15:46] time python2 fibo.py $n
1819143227

real	0m13.566s
user	0m13.432s
sys	0m0.020s
[fibo 15:46] time luajit fibo.lua $n
1819143227

real	0m1.028s
user	0m1.019s
sys	0m0.002s

but compiling fibo.c with -O3 makes it crazy fast:

$ gcc fibo.c -O3 -o fibo

$ time ./fibo $n
1819143227

real 0m0.077s
user 0m0.076s
sys 0m0.001s

While compiling with plan GCC gives 1.4 seconds, compared to luajit's 1.132