Bob and Gary,
The term 'update in place' means reading a specific portion of the media, modifying it, an re-writing back to the same place on the media. Back in the old days, you had to specify a specific cylinder, head, and sector where the original data was to be stored. If you added data at some later time (increased the size of the data; not just modifying existing data), you needed to write it to a different portion of the media, as the previous space was too small. Same with RW media. The sectors, if you will, cannot be increased in size, so an overflow area and appropriate links to it, are created so it appears as a contiguous data file. It's kind of like current hard drive technology - if the drive has defective cylinder, there are 'extra' cylinders which are automatically allocated, but this is transparent to the user; you don't know it's happening. Goes back to the premise that you can't put 1 gallon into a 1 quart jar.
And Bob, the buffer you referred to is not on the media, but in the computer memory. You can make any changes you want in the buffer, but it still must fit into the fixed sector size of the output media. Ned can probably back me up on this, but way back we had some programming languages that supported variable record lengths, but they were so difficult to manage that they died a cruel and almost instantaneous death - VTAM is one example.