[This is preliminary documentation and subject to change]
In dedicated application mode, any worker process can be configured to replace itself with another process. For example, if a worker process fails to respond to a ping in a certain amount of time, WAS marks that worker process as failing. Normally, WAS terminates that worker process and starts a replacement. If the Orphan Worker Process option is enabled, WAS leaves the failing worker process running, and starts another process in its place. This avoids service interruption of the requests coming in because they are simply cached until the new process is ready to process them.
Another benefit of placing worker processes in an orphaned state is to aid in debugging. WAS can be configured to run a command on the process (like attaching a debugger) when the process is placed in an orphaned state. With this feature, you can set up a test application pool that will launch a debugger against a worker process that fails to respond to a ping, thereby isolating it from its replacement.