Explicitly linearizing product of binary variables worsens performance

I am modeling a non-permutation flow shop model that includes sequence dependent setup times. As a part of the model, I get the non-linear expression

\[\sum_{j'\in\mathcal{J}}\sum_{j\in\mathcal{J}}x_{i,j',k}\cdot x_{i,j,k+1}\],

where \(x_{i,j',k}, x_{i,j,k+1}\) are binary variables. At first, I did not explicitly linearize this and just fed it to Gurobi's solver in the form of 

quicksum(x[i, j1, k] * x[i, j2, k + 1] for j1 in range(J) for j2 in range(J))

I then linearized this by introducing a binary variable \(z_{i,j',j,k,k+1} = x_{i,j',k}\cdot x_{i,j,k+1}\). This means that I added the following constraints:

\[\begin{align}
&z_{i,j',j,k,k+1}\leq x_{i,j',k}\\
&z_{i,j',j,k,k+1}\leq x_{i,j,k+1}\\
&z_{i,j',j,k,k+1}\geq x_{i,j',k} + x_{i,j,k+1}-1.
\end{align}\]

Instead of the non-linear expression, I then use

\[ \sum_{j'\in\mathcal{J}}\sum_{j\in\mathcal{J}}z_{i,j',j,k,k+1}, \]

and adjust the coding accordingly. 

However, I noticed that when doing this, the model performance worsens. The first and second log file of the model without and with (respectively) linearization is as follows: 

Nodes | Current Node | Objective Bounds | Work
Expl Unexpl | Obj Depth IntInf | Incumbent BestBd Gap | It/Node Time

0 0 23974.8571 0 129 68698.5714 23974.8571 65.1% - 0s
H 0 0 68683.571429 23974.8571 65.1% - 0s
H 0 0 68447.142857 23974.8571 65.0% - 0s
0 0 23974.8571 0 114 68447.1429 23974.8571 65.0% - 0s
H 0 0 59665.428571 23974.8571 59.8% - 0s
0 0 23974.8571 0 114 59665.4286 23974.8571 59.8% - 0s
0 0 23974.8571 0 110 59665.4286 23974.8571 59.8% - 0s
0 0 23974.8571 0 113 59665.4286 23974.8571 59.8% - 0s
H 0 0 54059.714286 23974.8571 55.7% - 0s
H 0 0 52184.714286 23974.8571 54.1% - 0s
H 0 0 51184.714286 23974.8571 53.2% - 0s
H 0 0 51084.714286 23974.8571 53.1% - 0s
0 0 23974.8571 0 115 51084.7143 23974.8571 53.1% - 0s
H 0 0 47952.000000 23974.8571 50.0% - 0s
0 0 23974.8571 0 115 47952.0000 23974.8571 50.0% - 0s
0 0 23974.8571 0 110 47952.0000 23974.8571 50.0% - 0s
0 0 23974.8571 0 110 47952.0000 23974.8571 50.0% - 0s
H 0 0 44996.285714 23974.8571 46.7% - 1s
0 0 23974.8571 0 110 44996.2857 23974.8571 46.7% - 1s
0 0 23974.8571 0 110 44996.2857 23974.8571 46.7% - 1s
H 0 0 44449.571428 23974.8571 46.1% - 1s
0 2 23974.8571 0 110 44449.5714 23974.8571 46.1% - 1s
H 33 36 44228.000000 23974.8571 45.8% 76.4 1s
H 36 36 44207.714286 23974.8571 45.8% 70.7 1s
H 62 74 43163.714286 23974.8571 44.5% 54.0 1s
H 96 104 42818.285714 23974.8571 44.0% 40.1 1s
H 131 149 42642.571429 23974.8571 43.8% 33.6 1s
H 151 160 42531.714286 23974.8571 43.6% 32.3 1s
H 154 160 40649.285714 23974.8571 41.0% 33.5 1s
H 259 273 40596.428571 23974.8571 40.9% 26.3 2s
H 283 288 40517.428571 23974.8571 40.8% 28.3 2s
H 289 288 40360.571429 23974.8571 40.6% 27.9 2s
H 366 442 40240.571429 23974.8571 40.4% 34.2 2s
H 494 521 40235.714286 23974.8571 40.4% 34.3 2s
H 2481 2424 40179.142857 23974.8571 40.3% 23.7 5s
H 2481 2423 40091.857143 23974.8571 40.2% 23.7 5s
H 2490 2423 40021.571429 23974.8571 40.1% 23.7 5s
H 2493 2303 40009.142857 23974.8571 40.1% 23.7 5s
H 2495 2190 39996.571429 23974.8571 40.1% 23.7 6s
H 2495 2080 39801.571429 23974.8571 39.8% 23.7 6s
H 2496 1976 39769.142857 23974.8571 39.7% 23.7 6s
H 2499 1879 39760.714286 24086.7143 39.4% 23.6 7s
H 2499 1785 39746.285714 24086.7143 39.4% 23.6 7s
H 2518 1708 39681.285714 25645.7432 35.4% 23.5 8s
H 2529 1628 39283.571429 25911.3138 34.0% 23.4 8s
H 2535 1550 39201.857143 25927.7095 33.9% 23.3 9s
H 2537 1473 39184.714286 25962.4714 33.7% 23.3 9s
H 2549 1406 39169.714286 25970.7721 33.7% 23.2 9s
2556 1410 25978.0548 36 133 39169.7143 25978.0548 33.7% 23.1 10s
H 2569 1347 38555.142857 25992.3124 32.6% 23.0 10s
H 2595 1297 38549.428571 25997.5487 32.6% 33.3 13s
2609 1306 27810.8004 223 131 38549.4286 27810.8004 27.9% 33.1 15s
H 2610 1242 38548.571429 27887.9543 27.7% 33.1 15s
H 2611 1179 38542.571429 27887.9543 27.6% 33.1 15s
H 2642 1139 38541.428571 28303.7403 26.6% 32.7 17s

Root relaxation: objective 2.397486e+04, 780 iterations, 0.03 seconds

Nodes | Current Node | Objective Bounds | Work
Expl Unexpl | Obj Depth IntInf | Incumbent BestBd Gap | It/Node Time

0 0 23974.8571 0 130 68698.5714 23974.8571 65.1% - 0s
H 0 0 66691.000000 23974.8571 64.1% - 0s
H 0 0 66651.000000 23974.8571 64.0% - 0s
H 0 0 66451.000000 23974.8571 63.9% - 0s
H 0 0 66371.000000 23974.8571 63.9% - 0s
H 0 0 65691.000000 23974.8571 63.5% - 0s
0 0 23974.8571 0 116 65691.0000 23974.8571 63.5% - 0s
H 0 0 59665.428571 23974.8571 59.8% - 0s
0 0 23974.8571 0 116 59665.4286 23974.8571 59.8% - 0s
0 0 23974.8571 0 109 59665.4286 23974.8571 59.8% - 0s
H 0 0 58221.000000 23974.8571 58.8% - 0s
H 0 0 55868.857143 23974.8571 57.1% - 0s
0 0 23974.8571 0 107 55868.8571 23974.8571 57.1% - 0s
H 0 0 54881.428571 23974.8571 56.3% - 0s
0 0 23974.8571 0 112 54881.4286 23974.8571 56.3% - 0s
H 0 0 49235.428571 23974.8571 51.3% - 0s
0 0 23974.8571 0 107 49235.4286 23974.8571 51.3% - 0s
0 0 23974.8571 0 110 49235.4286 23974.8571 51.3% - 0s
H 0 0 48931.714286 23974.8571 51.0% - 0s
H 0 0 48856.857143 23974.8571 50.9% - 0s
H 0 0 48522.571429 23974.8571 50.6% - 1s
H 0 0 47162.571429 23974.8571 49.2% - 1s
0 0 23974.8571 0 109 47162.5714 23974.8571 49.2% - 1s
0 0 23974.8571 0 109 47162.5714 23974.8571 49.2% - 1s
H 0 0 46912.285714 23974.8571 48.9% - 1s
0 2 23974.8571 0 109 46912.2857 23974.8571 48.9% - 1s
H 35 39 46852.285714 23974.8571 48.8% 32.2 1s
H 63 79 46758.857143 23974.8571 48.7% 20.8 1s
H 78 79 46570.285714 23974.8571 48.5% 17.8 1s
H 111 114 46169.857143 23974.8571 48.1% 14.6 1s
H 113 114 44275.285714 23974.8571 45.9% 14.7 1s
H 144 144 44246.428571 23974.8571 45.8% 14.6 2s
H 750 715 44151.428571 23974.8571 45.7% 12.8 2s
H 965 921 43984.285714 23974.8571 45.5% 13.2 2s
H 971 921 43841.428571 23974.8571 45.3% 13.2 2s
H 1025 966 43503.857143 23974.8571 44.9% 13.2 3s
H 1029 919 43473.857143 23974.8571 44.9% 13.2 3s
H 1031 875 43472.142857 23974.8571 44.9% 13.1 4s
H 1032 831 43443.857143 23974.8571 44.8% 13.1 4s
H 1033 790 43423.857143 23974.8571 44.8% 13.1 4s
H 1034 752 43418.000000 23974.8571 44.8% 13.1 4s
H 1034 714 43417.995621 23974.8571 44.8% 13.1 4s
H 1035 678 43371.428571 23974.8571 44.7% 13.1 4s
H 1035 644 43332.285714 23974.8571 44.7% 13.1 4s
1036 645 27376.5616 35 117 43332.2857 23974.8571 44.7% 13.1 5s
H 1038 613 43331.428571 23974.8571 44.7% 13.1 5s
H 1041 584 43298.428571 23974.8571 44.6% 13.0 5s
H 1042 556 43009.571429 23974.8571 44.3% 13.0 5s
H 1046 533 42400.714286 23974.8571 43.5% 23.9 6s
H 1076 523 42364.428571 23974.8571 43.4% 24.3 6s
H 1078 496 42080.285714 23974.8571 43.0% 24.3 6s
H 1108 500 41820.285714 23974.8571 42.7% 23.8 7s
H 1123 478 41780.285714 23974.8571 42.6% 23.6 7s
H 1124 456 41772.571429 23974.8571 42.6% 23.6 7s
H 1126 434 41746.571429 23974.8571 42.6% 23.6 7s
H 1127 415 41064.571429 23974.8571 41.6% 23.6 7s
H 1148 426 40855.000000 23974.8571 41.3% 23.3 8s
H 2981 1411 40746.285714 23974.8571 41.2% 16.6 10s
H 2982 1400 40515.000000 23974.8571 40.8% 16.6 10s
H 3761 1907 40500.000000 23974.8571 40.8% 15.8 11s
H 3804 1901 40416.714286 23974.8571 40.7% 15.7 11s
H 5069 2765 40384.285714 23974.8571 40.6% 14.7 11s
H 6173 3568 40354.285714 23974.8571 40.6% 14.1 12s
H 7792 4496 40334.285714 23974.8571 40.6% 13.1 12s
H 8234 4771 40154.000000 23974.8571 40.3% 13.0 13s
H 8236 4768 40087.142857 23974.8571 40.2% 13.0 13s
11976 7382 26876.7143 40 106 40087.1429 24002.7143 40.1% 12.1 15s
H13413 7831 39790.714286 24003.2289 39.7% 11.8 16s
H13417 7825 39635.571429 24003.2289 39.4% 11.8 16s
H14438 8414 39366.142857 24025.7143 39.0% 11.7 19s
H14439 8409 39302.570751 24025.7143 38.9% 11.7 19s
H14440 8408 39287.000000 24025.7143 38.8% 11.7 19s
H14443 8400 39144.428571 24025.7143 38.6% 11.7 19s
14446 9921 25526.1325 32 105 39144.4286 24025.7143 38.6% 11.7 23s
H15437 9919 39124.428571 24025.7143 38.6% 11.6 23s
18663 12006 28360.0035 62 103 39124.4286 24067.8571 38.5% 11.4 25s
H23064 13164 39069.714286 24124.8686 38.3% 11.2 27s

As one can see, after 27s, the obtained gap is 38% for the explicit linearization model, whereas this was 27% after only 17s for the model without explicit linearization. 

Question: What could be the reason for this? Is it because Gurobi uses some more efficient tricks to linearize products than the ones that I applied?