My most recent project, keysafe, is AGPLed, and the client and server are part of the same program, and a large amount of code is reused between them including http request generation/parsing, object queuing/storage, proof of work generation/checking, data types, and serialization.

So, I hope that there's not really a good reason to avoid using AGPLed code if it could end up running on the client. That would make development significantly less efficient when there's so much opportunity for server and client to share code.

Or it would limit the AGPLed parts to such a thin shell around the shared code that it might not be much of a barrier to creating a non-AGPLed server implementation. If keysafe's shared code was GPLed and only the server-specific code were AGPLed, a simple non-AGPLed server using the GPLed code be implemented by writing only 2 trivial functions; around 10 lines of code.