I found out that queries can be batched together to reduce the delay:

banshee --query-title --query-artist --query-duration

But still, I do not understand where the 200 ms overhead comes from.

Thank you for your bug report.

When you run "banshee" with some --query* parameter, it starts a new banshee process, which uses DBus to detect that there's already a Banshee process running, and then sends the query through DBus and displays the response.

So the delay might just be the startup of the new process (mono does add a slight overhead on startup).
I also get about 200ms response times on my system.

You might be able to do what you want by using DBus directly.
If your scripts are shell scripts, maybe with command-line tools "dbus-send" or "gdbus call ..."

As we implement the standard MPRIS interface, it would even work with other players.

For example, here's what I get:
$ time gdbus call -e -d org.mpris.MediaPlayer2.banshee -o /org/mpris/MediaPlayer2 -m org.mpris.MediaPlayer2.Player.GetMetadata
({'mpris:trackid': <objectpath '/org/bansheeproject/Banshee/Track/4314043'>, 'mpris:length': <int64 133982000>, ...},)

real	0m0.015s
user	0m0.000s
sys	0m0.004s

Thanks a lot! Now the response time is down to 20ms, that's much better. I'm closing the issue, then.