There are a number of things you can do to make sure your cached scripts run fast. The code for include/require looks like this: op2.u.constant.value.lval, inc_filename);]]> That is, we simply call the compiler which in an opcode cached architecture will be intercepted and the opcodes will be pulled from shared memory and fed directly to the executor. For the include_once case the code looks like this: value.str.val, &file_handle)) { if(!file_handle.opened_path) { file_handle.opened_path = estrndup(inc_filename->value.str.val, inc_filename->value.str.len); } if(zend_hash_add(&EG(included_files), file_handle.opened_path, strlen(file_handle.opened_path)+1, (void *)&dummy, sizeof(int), NULL)==SUCCESS) { op_array = zend_compile_file(&file_handle, ZEND_INCLUDE); zend_destroy_file_handle(&file_handle); } else { zend_file_handle_dtor(&file_handle); failure_retval=1; } } else { if(opline->op2.u.constant.value.lval==ZEND_INCLUDE_ONCE) { zend_message_dispatcher(ZMSG_FAILED_INCLUDE_FOPEN, inc_filename->value.str.val); } else { zend_message_dispatcher(ZMSG_FAILED_REQUIRE_FOPEN, inc_filename->value.str.val); } }?>]]> This rather complex-looking code makes sense in a non-opcode cached architecture because in order to figure out if we have included this file already we need to go through all the same steps as we do when we subsequently need to open the file, so it makes sense to just go ahead an open in it and then feed the opened file handle to the compiler. This saves a lot of work, but unfortunately it makes very little sense when we plug in an opcode cache.