This is indeed correct. The solution would be to store a timestamp
with each fragment group and have them "time out", and also to raise
the number of fragment groups. Unfortunately it's quite a large fix;
and since the impact is minimal (sending fragmented unreliable
messages isn't really a great idea in the first place) I will not fix
it for this version. I will however make sure it's fixed in the latest
release.
--michael