I found a curious thing happening depending on the delay between attempts. Here this is shown with a too-short file in all cases.
The first run is the 898cs we saw earlier. This was after a 5 second delay since the previous disc operation.
Next after a 250cs delay, when the drive has stopped spinning, the timing differs in the opening and writing stages, but the overall time remains the same. It's as if some sort of caching process is going on which causes the file to be opened more quickly, but somehow for the time saved to be used during the writing stage.
The final run here commences immediately, with the disc still spinning, and so we see an overall reduction.