Usually, I'd answer that TBB is designed for numeric computations and thus it does not work well with blocking I/O calls since it restricts the number of threads in order to avoid oversubscription. With modern asynchronous I/O frameworks, it might still be useful in other languages and for other applications.. but the nature of Python's web applications, I assume, implies that any computations are rather short and rely on Python interpreter often and heavily. Here, Python's GIL becomes a bottleneck, preventing applications from scaling and rendering TBB pretty useless. If these frameworks could run on multiple interpreter instances so that each thread would run a separate Python interpreter, it might make sense to revisit this conclusion.
