There is something unusual going on. Even if I reduce the number of runs to just 1, the total time taken is still close to 100 secs. I ran the script in Jupyter 5 times without changing any thing in the script and the same result was obtained.
The differential equations are the same you use in your benchmarks under the name three-body orbits. Integration results are correct. Fortran timing results are also from Verner9 method. Here is my fortran library with which i am comparing: https://github.com/princemahajan/FLINT