The benchmark is really crappy. If the language uses O(n+m) concatenation for strings of length n and m, then the total runtime would be O(n^2). That would happen if strings are immutable and no "stringbuilder" is being used. If that special case was optimized or if the strings are mutable, total runtime is O(n).

Here is the corrected C version:

    #include <stdio.h>
    #include <string.h>
    #include <stdlib.h>
    int main (int argc, char const *argv[])
    {
      char *str = malloc(sizeof(char) * 238890);
      char buf[5];
      int i, n = 0, len;

      for (i = 0; i < 50000; ++i) {
        len = sprintf(buf, "%d", i);
        strcat(str + n, buf);
        n += len;
      }
      return 0;
    }

It runs 100 times faster than the original program.

Edit: If you think I "cheated" by doing some calculations that I shouldn't do, here is a C++ version that runs just as fast as the improved C version:

    #include <string>
    #include <cstdio>

    int main()
    {
        std::string str;
        char buf[6];

        for (int i = 0; i < 50000; ++i)
        {
            sprintf(buf, "%d", i);
            str += buf;
        }
    }