As someone in that thread pointed out, I would say that the onus should be on the OS SDK to provide memory management primitives specifically for the purpose of security that can be zeroed out after use.
They also point out that the threat profile is someone who can read your system memory when 1P is locked (serious compromise already), but is unable to read it when it's unlocked (otherwise why bother reading it when locked), which is a pretty small set of use cases.