Exception handling, APM and remoting
Both remoting framework and use of APM produce a similar scenario: code you didn't write executes on a random pool thread without a thread exception handler.
When environmental factors pull the rug out from under this code you didn't write, running on another thread that doesn't have an exception handler, what we have is an unhandled exception.
When this happens in a Windows Service, a very common host for remoting, what we have is a whole new meaning for "unhandled" - not to mention a high speed way to terminate the service.
When the code is my own I install an exception handler. Framework code affords no such opportunity. But the framework itself could install an exception handler, if the thread lacks one. There's still the question of what to do with the exception, but I rather think the answer is obvious: log it, along with a stacktrace, in the application eventlog.