Hey,
there’s another (hard to find) thread with a similar problem, and some explanation in this and this issue.
In short, this is not a problem with the reachability analysis, but with the fact that these results (basically a point cloud) have to be translated to polygons somehow.
This will for example lead to deadlocks being seemingly excluded from the isochrone, e.g. in the northwest corner of your example.
If you want more detailed isochrones, have a look at the isochrone smoothing factor (not available via qgis).
Best regards