I don't think any arguments here on TMC would be convincing to the point where people can quantitatively prove one way gets better ROI than the other. Only Tesla would know that, and only Tesla knows when they will build the next GGF and Semi/MY production lines, what the CapEx will look like when that happens.
I can see a couple arguments for refresh:
1) refresh will likely cost much less CapEx than building new line, so if demand of S/X start to drop, especially S, because of people going to M3 (if M3 availability improves), then refresh could be a cheaper way to get more revenue coming in, also much quicker than building new production lines for MY/semi.
2) some refresh could improve economy of scale with M3, such as 2170 cell format and pack/module design, maybe even switching the motor from AC induction to PM