Here EC really shines. With -Dpreview_mt -Dec -Dmt it gets 1.8 req/s, while with just -Dpreview_mt it's less than 0.5. This is expected: if two CPU heavy tasks end in the same thread, since there's no snatching of tasks, it swamps the thread that then can't take further tasks.
Even if both CPU heavy tasks only block one thread, preview_mt keeps to round-robin spawn fibers to each threads, so 1/N of them will be sent to the blocked thread.
Here EC really shines. With
-Dpreview_mt -Dec -Dmt
it gets 1.8 req/s, while with just-Dpreview_mt
it's less than 0.5. This is expected: if two CPU heavy tasks end in the same thread, since there's no snatching of tasks, it swamps the thread that then can't take further tasks.