That's kind of a moot point when your browser handles most of your online interactions (and therefore a good chunk of your online identity, which is quite valuable to most people). Even if you isolate it as much as you can, which is a good thing to do in any case, it can still do a lot of damage without escaping the sandbox.
Yes, usage of isolated browser instances should be restricted to information within a single context or risk profile. E.g. a stateless, frequently rebooted VM for occasional use of a particular extension. Or a Bromium micro-VM for each tab, redirect, etc.
As for practicality, if your daily workflow involves a browser extension that has no replacement, the options are:
- stop doing the task
- all browsing with insecure browser, no isolation
- single task with insecure browser, no isolation
- single task with insecure browser, some isolation
Most people will do #2 or #3. Those who care about security will do #4, with quality of isolation dependent on their threat model.