Would like to know if Streamlit is following up with any plans to improve the running speed?
If we have any ml or dl model the we have st.cache to store the modle in cache which helps to avoid loading modle multiple times.
I mainly look at the user experience based on multi page development. Perhaps I can ask a different question, whether streamlit can support local loading. Currently, the form of global loading causes a global refresh of the entire page every time, and I feel that it still lacks some continuity and smoothness in use. 
You might look at using st.fragment - Streamlit Docs