Thanks Krzysz! That fixed it and now CLI rendering of a file with a tremendous amount of arcs seems to be a bit faster:
Unpatched trunk: 0m9.161s
Patched trunk: 0m7.577s

However, something else I might have to investigate is that with my sample file, experimental seems to have suffered a serious performance regression:
Unpatched experimental: 0m20.539s
Patched experimental: 0m19.482s