If the community discovers after the fact that an existing SRFI's scope
is too broad, it seems it would be best to make a new SRFI (with a new
SRFI number) that is simply a subset of the old SRFI. This has been done
in the past: all the SRFIs with "reduced" in their title.
All four out of over 200, none of which represent splits. 13 -> 152 (strings) and 114 -> 128 (comparators) represent genuine simplifications, though some procedures from 114 were added back by 162. 99 -> 131 (R6RS-ish records) represents a reduction to those features of 99 that can be portably implemented, and 150 is a different approach to 131.
Scheme
implementations that ship the old SRFI could also ship the new one, with
the identifiers simply being aliases to the old library.
Quite so. But that's not the story here: there are no similar facilities for lexical syntax. I think it would be a WOMBAT (waste of money, brains, and time) to split 4/160 (heterogenous vectors) or 135 (texts). It's true that 163 (array literals) and 164 (arrays) are split, but we are explicitly told that 163 can be used with any array SRFI, though in fact the only implementation of either 163 or 164 is Kawa.