Speaking about Ethernet.
Those STM32s with integrated MAC controller still require an external PHY chip.
An alternative is to use W5500 MAC+PHY (works over SPI), and then use pretty much any MCU you like.
What is the point of being limited to MAC-enabled STs? Right, there is no point at all.
W5500 has its own built-in stack with socket API, alternatively you can run a stack on the main MCU and use W5500 just to send/receive frames.
Then, WiFi and Espressif. I don't get why people still mention, still talk, and still use ESP8266. Mental inertia, I suppose.
Practically, only ESP32 line should be considered.
And if we're talking about ESP32, a possible solution is to run the whole networking stuff solely on ESP32. Both Ethernet and/or WiFi and/or cellular. Then, ESP32 could provide a simple serial interface to the main MCU. One possible implementation of that scenario - a product of my company,
https://vcon.io