Performance difference of loc method between Pandas 1.3.5 and Pandas 1.4.2

I have a question regarding the performance of the loc method in Pandas.

I tried the following code and measured its time to execute:

import timeit
import pandas as pd

data = {"id": [str(i) for i in range(100_000)]}
data_frame = pd.DataFrame(data=data)

def test_loc_read():
    print(pd.__version__)

    data_frame["value"] = 0
    for label, _ in data_frame.iterrows():
        data_frame.loc[label, "value"] = 1

if __name__ == "__main__":
    print(timeit.timeit(test_loc_read, number=3))

It seems that the performance increased from Pandas 1.3.5 to Pandas 1.4.2.

I measured approx. 14 sec. with Pandas 1.4.2 and approx. 50 sec. with Pandas 1.3.5.

Can somebody explain where this performance gain is coming from? I already checked the Release Notes but couldn't figure it out.

Thanks!