I have the same situation with Reeder (described in the parallel thread).
It is indeed absolutely awful, but I'll let the developers know.
Sounds a bit like a chicken egg problem: 1000 unread articles are not really much these days with a decent number of feeds. Also memory (in VMs), memory management, request scalability and database backend shouldn't be that vulnerable to such requests these days. Nevertheless programmers should of course use the proper API calls to send proper requests if they are implemented in the API.