Same test: https://build.chromium.org/p/chromium.fyi/builders/Fuchsia/builds/8812

Inspecting the test code, it appears that it is making the assumption that all Chrome IPC is serialized via the IPC thread, so that if we post a Task to the IPC thread and signal an event, then when we see the event the IPC must have been received by the peer (see https://cs.chromium.org/chromium/src/ipc/ipc_sync_channel_unittest.cc?type=cs&l=1189).

Now that Chrome IPC runs on Mojo, the Mojo and Chrome IPC threads differ:
- Mojo's is a TestIOThread created at https://cs.chromium.org/chromium/src/ipc/run_all_unittests.cc?dr&l=21
- Chrome's is a member of the Worker test helper class (see https://cs.chromium.org/chromium/src/ipc/ipc_sync_channel_unittest.cc?gsn=ipc_thread&l=251).

So I think that this test is now inherently racey, and needs updating to post a message to the Worker ipc_thread_ that then posts a message to the (process-wide) Mojo IPC thread to signal the event.

Except that this is in-process IPC, which I think Mojo can shortcut to pass directly between threads?

Yeah, Mojo's IPC thread is never relevant for in-process message passing.

No sign of this flaking any more, so closing as obsolete.