The divisions between LC and PCC and within PCC on this question have so far not been bridged successfully, in part because the terms in which it is discussed have been problematic. As Ian's question indicates, there is a presumption that an authorized access point represents by default one work or one expression; but in the case of translations, LC and long-standing practice have preferred to let one established heading represent a variety of distinct expressions.
One solution would be to introduce a different default. If an expression AAP consisting of a work AAP plus a language were understood to represent by default a class of expressions and not a single expression, then libraries that prefer to extend the class-of-expressions AAP with more specific qualifiers, e.g., translator names, could do so without calling into question the assignment of the unqualified AAP. The class designation would still be true. The decision about whether to collocate entries with a class-of-expressions AAP or with something more specific to a particular expression or with some hybrid of these two practices could be made locally and case by case and would not pose the kind of dilemmas which the current state of affairs, where all of these AAPs are presumed to represent particular expressions, inevitably does.
A class-focused AAP would also be a more persuasive model for conventional uniform titles and selections, additional cases where the same AAP may represent collections with markedly different contents. Different libraries would be able to make different decisions about how finely to distinguish aggregate works and expressions falling into such classes; and cases where greater specificity is desired would not invalidate a decision to describe resources only in terms of class AAPs.
And because Ian mentioned the "undifferentiated"--we need to keep in mind that the undifferentiation that exists among the members of a homogeneous class is different from that for members of a heterogeneous class. The English translations of the Heart Sutra collectively will have a lot in common, where the works of several different persons whose names all fall under an undifferentiated name AAP most likely will not. It makes sense that we would embrace the former and avoid the latter when constructing AAPs to provide users with meaningful collocations of resources.