Having done a bit more reading, I think the status of using punched cards for program data was very much tied to the evolution of operating systems for the #IBM7094 and related machines. I get the impression that punched cards were very much an #IBM thing, users never really liked using them, and that people abandonded them as soon as they could. I don't think they scaled well with more complex site operations and/or faster machines.
#FORTRAN II was released in 1958. It definitely eventually had a stand-alone monitor program (essentially an OS), and I have no reason not to believe it was available from the start, so I suspect this is how Mathews would have at least started work on #musicn. The FORTRAN monitor has clear support for data cards - there's a job control command to mark the start of data sections and the I/O library has support for on-line card reading.
By the time you get to 1960, #IBSYS (a later IBM operating system) has been released. This encapsulates the FORTRAN monitor, but the documentation is clear that on-line card reading via the FORTRAN I/O libraries is not available under IBSYS. It's likely that there are other libraries which do support it, but presumably this would require user code changes. I think sites were starting to switch to off-line translation of cards to magnetic tape by then anyway, although I'm not sure exactly how that would have affected programming.
By the time you get to #CTSS end users are at least strongly encouraged to use the disk filesystem for on-line applications, with operators copying disk files to and from cards or tapes in batch jobs as required. The CTSS system libraries have routines for using disk files as virtual tapes. CTSS was 1961, so I find it really impressive that this transition happened over only three years.
1/n
#retrocomputing #synthesizers