Efficiently pass numpy array to Matlab function

I am working with some python codes that call a Matlab function. The Matlab function takes a 3D array as input and the size is about 87*195*126. I modified the mlarray_sequence.py as suggested in this post. The time spent on converting to Matlab double array does significantly reduce (~0.4 sec). However, it still seems that it takes about 10 seconds to pass the array to the Matlab function, as can be seen by timing each step as follows:

import numpy as np
import matlab.engine
import matlab
from io import StringIO
from os.path import abspath, dirname, join
import timeit

def test1():
    start_time = timeit.default_timer()
    matlabEngine = matlab.engine.connect_matlab()
    directory = abspath(dirname(__file__))
    matlabFolder = directory
    matlabEngine.addpath(matlabFolder)
    matlabEngine.cd(matlabFolder)

    timepoint = timeit.default_timer()
    print('Elapsed time for connecting matlab is {} sec'.format(timepoint - start_time))

    img = np.random.rand(87, 195, 126)
    timepoint1 = timeit.default_timer()
    imgMatlab = matlab.double(img)
    timepoint2 = timeit.default_timer()
    print('Elapsed time for creating matlab array is {} sec'.format(timepoint2 - timepoint1))

    out = StringIO()
    err = StringIO()
    matlabEngine.test2(imgMatlab, stdout=out, stderr=err)
    print(out.getvalue())
    print(err.getvalue())
    timepoint3 = timeit.default_timer()
    print('Elapsed time for matlab script is {} sec'.format(timepoint3 - timepoint2))

test1()

function result = test2(img)

tic
result = 1;
toc

And the output is:

Elapsed time for connecting matlab is 0.0893801 sec
Elapsed time for creating matlab array is 0.40231419999999996 sec
Elapsed time is 0.000603 seconds.
Elapsed time for matlab script is 10.4323424 sec

[Done] exited with code=0 in 11.491 seconds

So is there any way to reduce the time spent on passing the array to matlab function?

P.S.: I have already shared the matlab session by typing matlab.engine.shareEngine before I run the script.