> (1) It would be one more convention everyone has to agree to.

So are UUIDs.

> (2) It has no benefit because the uid is hidden from the user anyway.

Neutral.

> (3) It clashes with the earlier convention of record-name-UUID proposed
> in R6RS.

This is a good argument.

> (4) It breaks down should one SRFI want to define two different record
> types with the same name (unlikely but I not completely).

In that case you would append a word to disambiguate them.

The major reason against I tend to disfavor UUIDs is that they are not
human-readable, and it's best to avoid non-human-readable things in
source code. That means they are not self-documenting either. If we use
UUIDs for standard record types, it would be prudent to keep a list of
them anyway. Our most important goal is to avoid name clashes, which any
of the proposals so far should be able to accomplish easily.