There was a bug in GLOM that I had to revisit, and when I did revisit it... this made me question the optimization. The important optimization was avoiding creating new identities of empty blocks. But once you create an identity for a block holding an element, that's the only identity you need in the average case...regardless of how many elements you add.
So it was "avoiding one Stub allocation" (8 platform pointers) vs. having to run many cycles of interpreter code to branch the path to handle the QUOTED! case. Not worth it--at all.
That aside: I think that considering QUOTED! as a "container" is likely misguided in general. What quoting "means" is more likely to be dialected as some kind of marker. I've begun to think that if anything, single-element SPLICE! is better... in particular because you don't have to do "mental math" to subtract a quote level from "what you're actually talking about" (''2 as a container for '2 isn't as readable as ~['2]~) And yet I'm still not sure if FOR-EACH should even allow that--so quoted is down the rankings a lot.
So I'm now pretty doubtful that things like FOR-EACH should iterate QUOTED! as if it were a "container of one item" as some kind of optimization. Maybe internal code can use the quote byte creatively here or there, but exposing this to user is a false economy--as I say, usermode code can't capitalize on the benefit because every bit of branch code in usermode costs more than the savings could be.