The way disk i/o in unix works is that data is parked in an in-memory cache in the kernel - it is not guaranteed to be written to disk when the write() system call is invoked. Every 30 seconds or so the syncer daemon issues a sync command. This forces the kernel to write everything in the kernel buffer to disk.
What you are seeing is an incompleted write operation - for whatever reason. Common reasons are - a signal was sent to the process that terminated it, write() or sync failed because something else filled up the
disk (maybe a temp file) and then that file went away, disk errors caused a fatal error. If it's an nsf mounted disk then the network also becomes an issue. What errors do you see in the log?