>From my understanding there are a few issues being discussed here
(with possibly overlapping consequences) -- O(1) performance of
strings and using strings for octet sequences.

SRFI-4 defines homogenous numeric vectors which seem like they would
suffice for implementing octet streams independant of strings and, if
not, a few Scheme implementations (larceny, MIT/GNU-scheme) provide
byte-vectors.

As for O(1) performance, there's undoubtably a need for
non-O(1)-string-like-heavy-weight-objects (herein called ropes) which
some implementations choose to implement and call them strings (as
permitted by r5rs).  The current proposal seems to urge
implementations to provide these as a seperate data type.

Two reasons for this distinction seem to be:

  ~ efficiency

    Carrying around the full weight of a rope for all strings might be
    too excessive, but then r5rs doesn't define the concept of a
    fixnum which most implementations utilize to avoid needing to
    provide heap allocated bignums.

  ~ use in binary I/O

    relating back to the discussion above -- should an alternative
    idiom be employed for this rather than reusing strings for this
    purpose.

-jivera