From David Holmes:

Re: "Maybe the logic for "remapping the RO section to read-write" is faulty on Windows and in rare occasions would unmap the RO section"

That rings a bell:

// Remap a block of memory.
char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
                       char *addr, size_t bytes, bool read_only,
                       bool allow_exec) {
  // This OS does not allow existing memory maps to be remapped so we
  // have to unmap the memory before we remap it.
  if (!os::unmap_memory(addr, bytes)) {
    return NULL;
  }

  // There is a very small theoretical window between the unmap_memory()
  // call above and the map_memory() call below where a thread in native
  // code may be able to access an address that is no longer mapped.

  return os::map_memory(fd, file_name, file_offset, addr, bytes,
           read_only, allow_exec);
}

What happens to the memory contents between the unmap and map?

Change below to use VirtualProtect got GetLastError = 75

http://cr.openjdk.java.net/~coleenp/8087153.01/