Okay so, profiling the function shows that 84% of the time is spent at line 375 of tropical_cyclone.py:
haz = cls.concat(tc_haz_list) . In that function, 100% of the time is spend on line 1056: haz_concat.append(*haz_list). @emanuel-schmid did the appending of the hazard changed between 3.0.0 and 5.0.0 ?

Note: The profile results are run only with:

tracks_SI_proba = TCTracks.from_ibtracs_netcdf(year_range=(1990, 2023), basin='SI')

with out the following code, as it takes 1h per run... I will run the profiling for the slow version as well. But I would assume, maybe wrongly, that the behaviour will be the same.

tracks_SI_proba.equal_timestep(time_step_h=1)
tracks_SI_proba.calc_perturbed_trajectories(nb_synth_tracks=9, max_shift_ini=0.75, max_dspeed_rel=0.3)

@emanuel-schmid I can send you the profiling data per slack, github do not support the.Iprofformat

Update: haz = cls.concat(tc_haz_list) is also the slowest line with the perturbed trajectories.

One step further (without perturbed trajectoires), in hazard.append() the bottle neck is line 916 of base.py:
centroids = Centroids.union(*[haz.centroids for haz in haz_list]). Here, is Centroid.union which takes time
--> centroids.append(cent) --> self.gdf = pd.concat([self.gdf, centr.gdf]) is the guilty one.
(At least without perturbed trajectoires)

climada/hazard/centroids/centr.py line 347:

    def append(self, centr):
        """Append Centroids

        Note that the result might contain duplicate points if the object to append has an overlap
        with the current object.

        Parameters
        ----------
        centr : Centroids
            Centroids to append. The centroids need to have the same CRS.

        Raises
        ------
        ValueError

        See Also
        --------
        union : Union of Centroid objects.
        remove_duplicate_points : Remove duplicate points in a Centroids object.
        """
        if not u_coord.equal_crs(self.crs, centr.crs):
            raise ValueError(
                f"The given centroids use different CRS: {self.crs}, {centr.crs}. "
                "The centroids are incompatible and cannot be concatenated."
            )
        self.gdf = pd.concat([self.gdf, centr.gdf])

I think I found a potential solution, I'll make a PR 🙃