I don’t do this and it seems like a hassle but it is an interesting question: would this be more or less private than a single device?

One thing I can think of is most tracking algorithms probably assume each user has a single cell phone (either explicitly or the ML data is biased in that direction). So splitting your time across two devices probably messes with whatever user-behavior buckets they place you in. They might think you’re two people in the same household, for example.

It's actually more in-line with how real world threat models work. You always operate under the assumption that all of your hardware is compromised, then build layers of trust around that to determine which device should be used when. Not super practical for an end-user, but it's definitely better than having a single device that you always second-guess.

Maybe with different VPN endpoints on each device it could work. But it would require a strict discipline. Google is very good at finding the same person in 2 very different devices.