Is that with the 900 Vcal? Looking at your circuit I think the input impedance on the ADCs is way too high, especially the 470k on the VT circuit. 12K on the CT circuits is in the ballpark (and so is the ICAL) It doesn’t give the sample and hold capacitors enough current to charge. There are a few things that I do differently with MCP3208s to make them work.
When I was using voltage dividers successfully (I use op amps now), I was using 4.7K resistors, with a 10uf capacitor where you have the .1uf cap. I know that works pretty well, but at anything above about 500Khz on the SPI clock, you need to increase the sample and hold time. I developed a technique to do that and I’ll post a link to that later.
I used that circuit on both the VT and CT inputs with success.
The tell tale that the ADCs are not charging to full voltage is that the offset values sag. If you print the offsets out after the thing settles down, they should be around 2048. A much lower value would be consistent with insufficient charging and your need to multiply the result with that large Vcal.
The Arduino ADCs are not as demanding as the MCP3208, but they are also slower. That explains why it works with them.
You should be able to run the ADCs at 1Mhz if you use the S&H lengthening technique. I run them at 2Mhz with the op-amps (very low impedance).
Here’s the link to the old post on increasing the ADC S&H time:
https://community.openenergymonitor.org/t/esp8266-wifi-power-monitor/1692/19?u=overeasy