[FabHK]

Makes sense, thanks. I think my mental threat model was different:

you're talking about dodgy (potentially compromised) software: better run in the browser sandbox (albeit imperfect) than natively.

I was thinking of sensitive software (eg secret chats) I want to protect from attack: better run natively than in the (imperfect) browser sandbox.

In the context of this discussion (Slack), your threat model probably makes more sense.