Filesystem IO through DOS APIs does not have bounded execution time by design (it can even cause user interaction, ie. the infamous "Abort, Retry, Ignore?"). On the other hand the timing is pretty consistent in non-error situations.

Also synchronous (and also often non-preemptible) disk/FS IO was for the long time the only way it was done. For example the reason that on Linux process that got stuck accessing NFS or removable drive that got removed was unkillable (in 'D' state) was to preserve illusion of such semantics to user processes (ie. that IO to filesystem and block devices will never fail with EINTR). And I would not be surprised if many deeply-embedded RTOSes do the same to this day.