The main motivation is that the new failure code is written in Rust so calling it directly makes more sense than indirecting through native runtime code.

This also lets us to have fewer stack-switching mechanisms as upcalls are dealt with differently than other foreign functions.