In case Mr. Roewen's reply was too terse or if you don't know much
about virtual machine implementation (not that I do...), here is a
fuller explanation of why the interepreter can be faster in GCC:

In GCC, unlike standard C, you can take the address of a goto label as
described here:

http://gcc.gnu.org/onlinedocs/gcc-4.0.2/gcc/Labels-as-Values.html#Labels-as-Values

So whereas normally in a C implementation you would do something like this:

start:
switch(*ptr) {
  case INSTRUCTION_1:
    do_stuff();
    ptr++;
    goto start;
  case INSTRUCTION_2:
    do_stuff2();
    ptr++;
    goto start;
  /* ... etc ... */
}

That is, each instruction for the virtual machine is represented as an
otherwise meaningless integer and you have to do a switch statement
for each one, and however your C compiler optimizes that, is what you
get. On the other hand, if you can take the address of a label, you
can go through the virtual machine instructions and replace each
arbitrary integer instruction with the address of a goto label
corresponding to that instruction. Then you can do this:

instruction_1:
  do_stuff();
  ptr++;
  goto *ptr;
instruction_2:
  do_stuff2();
  ptr++;
  goto *ptr;
  /* ... etc ... */

Since this provides a speedup for every VM instruction, it's quite a boost.

I think it's rather unfortunate this is not part of standard C.