* Internal: Reorganize code protecting s7 gc
  variables a little bit. Might have fixed some
  potential problems.
* Internal: Preallocate a little bit of memory for
  vectors. Also remove some thread assertion code
  from non-debug builds. Might improve performance
  a little bit.
* Internal: Make sure memory allocated by the RT
  memory allocator is always aligned at maximum.
  This could theoretically increase performance.