Ray ActorPool with 2 actors for Tensorflow resent-50 prediction is not performance better than single actor pool

Ray Core
1 
import numpy as np
import sys
from _predict import PredictService
from config import InternalConfig
from _backend import MongoORM
from logger import AppLogger
logger = AppLogger(__name__).get_logger()

import time
# num_cpus = psutil.cpu_count(logical=False)
num_cpus = len(psutil.Process().cpu_affinity())
print('Initializing Ray on {} cpus.'.format(num_cpus))
ray.init(num_cpus=num_cpus)
print('Ray initialization {} on {} cores.'.format(
    'successful' if ray.is_initialized() else 'failed', num_cpus))


@ray.remote
class MyActor(object):
    def __init__(self):
        import tensorflow as tf
        self._model = tf.keras.models.load_model(
            './assets/sample/models/resnet-50.h5')

    def func(self, x):
        logging.error(f'calculating {x}')
        out = self._model(np.zeros((10, 512, 512, 3)))
        return x * 2


actors = [MyActor.remote() for _ in range(2)]
actor_pool = ray.util.ActorPool(actors)


def ray_batch(args_list):
    res = actor_pool.map(lambda a, v: a.func.remote(v), args_list)
    print(list(res))


import time
st = time.time()

args_list = [1] * 10
ray_batch(args_list)
print(time.time() - st)

The performance gain is 5% here 2 ray actors vs 1 ray actor
im testing on 2 core machine (edited)