Why the IRFs are so abnormal? Is it either normal or not?

[jpfeifer]

That looks a lot better. The IRFs do return to the steady state, but they are very persistent. See also

Longlasting IRFs

That is hard to tell. Most of the times, we actually want out model to feature persistence, i.e. endogenous propagation. In a quarterly model, 50 periods is a bit more then 10 years. That can still be considered a business cycle frequency.