Higher expectations are the major issue: undo being the canonical example. The general problem of distributed system data synchronization is another: things tend to demo well enough on a local development instances, and then fall to pieces in the unreliable field. Showing synchronization status, merging changes in a semantically correct way, useful UX affordances around all this, etc. is all very difficult work.

Of course it all depends on the domain and architecture of the app in question, so counter-examples and exceptions are easy enough to find.

You always have the option of simplifying by writing the app in the "web style" where it simply refuses to run if it can't connect to the server (if there even IS a server).

Of course, people might indeed think that's lame and sucky whereas they'd forgive it for a web app. This is one reason I made my app visually look like a web app e.g. no menu bars and a bootstrap/web style button/popup aesthetic. If it looks like a duck, people are more forgiving if it quacks like one sometimes too :)