The loss did occur on simple ping commands, only on 2 out of 5 ports. The vendor confirmed the behavior to be faulty and took the switch back.
Maybe it was just a faulty model? However I do use multicast in my network (corosync) and a lot of 10G switches seem to have problems with that, maybe this was the case here, too.
The exact model is TRENDnet 5-Port 10G Switch, 5 x 10G RJ-45-Ports and there sure seem to be quite some people having issues as well...
Yes, some chips (or rather parts of all chips) are spare on enterprise SSDs. You can even see how much is left via smart data