A copy to/from would technically work since NVMe is full duplex, although I'm not sure how accurate it would be. Ideally you'd have one giant file/ISO to test rather than lots of small files. You're right, HDTach is limited, although it may still be useful as seen here; it would just take longer for the drop to occur. I'd have to extrapolate SLC cache size from the results which is a bit tricky but not impossible. Not sure on the limitations of HDTune's free version actually.