This is due to caching the resolution of GetType.GetTypeClass, which appears as identical Ident trees in the parent of Thing and in the RHS of e.

From the superclass context of Thing, it incorrectly cached at an outer context, where the lookup from e found it as a definition at higher precedence than the import.

A couple of notes to self or reminders:

The trees don't distinguish whether the ident is due to implicit search or import. The comment at typedIdent is false, because the elidable prefix does not result in a Select. That is, I'd expect the RHS of e to be a typed selection.

Scala 2 certainly transforms to a select:

Select(Ident(p.GetType), TermName("GetTypeClass"))

or -Yprint-trees:format

          ValDef( // val e: p.GetType[p.Class]
            0
            "e"
            <tpt> // tree.tpe=p.GetType[p.Class]
            "GetType"."GetTypeClass" // implicit val GetTypeClass: p.GetType[p.Class] in object GetType, tree.tpe=p.GetType[p.Class]
          )

This doesn't matter for Scala 2 -Wunused:imports because the typer records the imports it used for lookups.

Removing the loopOverPrefixes doesn't break any tests. Either that progressed elsewhere or a better test is needed. (I think I restored that code at some point last year, but I don't have commit history to see why.) (Edit: https://github.com/scala/scala3/pull/22751/files)