It's quite simple: write-behind caching means that after you have (for example) saved a file in a word processor, the saved file may live in the cache in RAM for a while before being written to the disk. If the computer crashes during that time, or is just switched off without using some command that flushes the cache to disk, then the file will be lost, or will still exist in its previous version. Worse still, if only some of the blocks of the file have been written, it will be a garbled mixture of the old and new versions.
Without caching (or with write-behind disabled) you can be sure that any saved file will have actually made it to the disk in a short time, so that waiting for a couple of seconds, or waiting for the disk light to stop flashing, before switching off the machine would ensure that the work was properly saved. With MS-DOS, it was normal to switch off the machine without any shutdown procedure.
As Stephen Kitt notes in another answer, SmartDrive would make sure the cache was flushed after exiting an application before the DOS prompt was shown again, making it safe to exit the program and then switch off. What wouldn't be safe would be to save a file and then switch the machine off without exiting the program. This behaviour of SmartDrive made it safer but a bit less effective, because a sequence of DOS commands (maybe from a batch file?) couldn't benefit from write caching between one command and the next.
The risk of crashes or unplanned shut-downs makes it safer to disable write-behind caching as the default setting, so that the cache is used only to hold copies of blocks on the disk that have been recently read or written. Even this use of a RAM cache is highly effective in speeding up average access times, because a typical workload would contain ten times as many block reads (for programs, input files, fonts, etc.) as block writes.