* map-ref: Should the SRFI guarantee that the success and failure
procedures will be called in tail positions? Would a similar
requirement make sense for some other procedures? (I did not
check.)

=> That's an interesting point. My sample implementation currently fails to tail-call them, but this could be amended. I quickly scanned over the specification, and the two procedures to map-ref are probably the only ones for which it makes sense to demand that they are being tail-called. I have to think about it. The equivalent hash-table-ref from SRFI 125 should also be looked at.