Unlike netCDF, reading binary data requires reading in the whole
grid at once. That's one of the big advantages of netCDF. So
you're not gaining anything with the repeat loop. To read it in
segments you would need to define a smaller grid to read a subset
of records and write that out, the do a second read, skipping
those first records [emphasis mie], and so on.
Aha! That's the trick I needed to know! In particular the "skipping" part didn't occur to me. Thank you for your help.
Extending this discussion . . . As you say, netCDF is a convenient format in many counts, obtaining slices of an array one of them. That's why I wanted to convert the binary file into netCDF. Ferret is the best tool to do so as far as I know. I wish ncgen accepted raw binary arrays for "data". Converting the double precision binary into ASCII, which ncgen requires, would more than double the size, not to lose precision.
Ryo