You only need to look at the second page of the data sheet:
Do you mean how long from calling one of the methods (calcIrms( ) or calcVI( )) to the method returning a result? That depends on what you ask it to do and how long it takes for the ADC to get a sample. The Atmel '328P is particularly good for its class in this respect at 104 µs (when free running). My measurements for the sampling rate of emonLib, measured with a standard emonTx V3.4 (and the emonPi is the same processor) are:
calcIrms( ) - approx 5588 current samples per second.
calcVI( ) - approx 2535 pairs of voltage & current samples per second.
(see Sampling rate of emonLib)
The data sheet (again 
) says
What I don’t know is what 3rd party libraries in the ESP might be doing in between emonLib and the ADC itself, so the sample rate you get will most likely be significantly less than this. This is certainly the case with the Arduino libraries as used in emonLib, as you can see from the numbers above. There is a faster way - the way we do it in emonLibCM and emonLibDB, where the ADC free runs and dumps the values into a memory location, then generates an interrupt to signal the main software loop to take the value and process it. Before you ask about converting either to run in the ESP32, it appears to be possible from what I read here esp32_technical_reference_manual_en.pdf, but I’m not going there.
Most people who use the ESP8266 and the ESP32 have a separate ADC and use the ESP to do the maths and send the results out - usually by Wi-Fi.