Just a useless test

try.cpp

#include <sys/time.h>
#include <stdlib.h>
#include <stdio.h>
#include <thread>
typedef struct quux {
    int i;
} quux_t;

void foo(int run) {
	quux_t *bar;
	printf("thread on\n");
	int i;
	for(i = 0; i < run; i++) {
	      bar   = (quux_t *)malloc(sizeof(quux_t));
	    free(bar);
	}
}
int main(int argc, char **argv) {
    struct timeval tv;
    gettimeofday(&tv, NULL);
    int i;
    long usecs_start = tv.tv_usec + tv.tv_sec * 1000000;

    int th = atoi(argv[2]);
    std::thread threads[th];
    for(i=0; i< th; i++){
	    threads[i] = std::thread(foo,atoi(argv[1]));
    }
    for(i=0; i< th; i++){
	    threads[i].join();
    }

    gettimeofday(&tv, NULL);
    long usecs_end = tv.tv_usec + tv.tv_sec * 1000000;
    printf("%ld\n", usecs_end - usecs_start);
    return 0;
}

Compile with

g++-5.2 0.cpp -O2 -Wno-unused-variable -Wno-unused-parameter -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free -std=c++14 -lpthread -ljemalloc

-Wno-unused-variable -Wno-unused-parameter are used to avoid the optimizer to just remove the unused result. resulting in light speed execution...

Non using jemalloc will result in 1sec slower execution

Cpu Amd A10 7400P @ 2.5Ghz; Linux 4.3. something

for i in {1..10} ; do ./a.out $N_OBJS 2; done|awk '{s += $1} END { print "avg: ", s/NR }'
avg: 345370

for i in {1..10} ; do ./a.out $N_OBJS 1; done|awk '{s += $1} END { print "avg: ", s/NR }'
avg: 344059

for i in {1..10} ; do ./a.out $N_OBJS 3; done|awk '{s += $1} END { print "avg: ", s/NR }'
avg: 355574

for i in {1..10} ; do ./a.out $N_OBJS 4; done|awk '{s += $1} END { print "avg: ", s/NR }'
avg: 380500

Of course after 4 thread time start to grow

-W... are for warning, not for compiler features. For compiler features you should use -f...