Actually I'm not real sure if I agree with you here. The arguments you have made don't gibe with my understanding of fixed width, and non fixed width. If the filesystem is doing transparent compression, then the file system is in fact a stream and you are adding a layer of work (no doubt handled by subsidary controllers or whatnot) to then represent the data at a logical fixed width format. Presumably if someone thought the cycles weren't worth wasting doing CSV handling (or XML for that matter) they could also offload the task to subsidiary hardware, and we are back at point one, which is that fixed width records dont scale well. :-) The only argument left for fixed width records is they dont care what value is within them, but this also goes for at least one form of delimited files (length delimited fields) which also scale well.

BTW, im not saying that fixed width doesnt suit some circumstances. Just that I dont think those circumstances are really that common. (These days. :-)