The WebSocket code in the Web Access server is written to use NumPy to accelerate the decoding of data. This is massively faster than trying to do this in pure Python code.
However we've disabled the support of NumPy (to avoid dependencies?) and will always uses the slower fallback.
Fortunately this only affects data sent from the browser to the server, which is generally not that much. But it will be an issue as we want more data to flow in that direction (e.g. microphone, large clipboard or file transfers).