These don't expose anything other than subpixels in fonts, which are already exposed. We bundle fonts and control what fallback is used per language, especially now win7/8* are dropped, I expect win + mac to very tight, and linux ALL fonts are bundled.
I enabled textmetrics on TB128 and tweaked my code locally to compared english to all other languages
tl;dr: ja, ko, zh-TW are an artifact of #42465. We disabled textmetrics in FF102 (? I think it was this release) since they were new and unlikely to be used or cause breakage. It's now 2+ years later. I see no need to keep them disabled
Note: these are the measurements of MY windows 11, and also note that the non-actualBounding tests are a single glyph. These are not super deep dives into entropy - e.g. alphabeticBaseline is 0 for everything (I would have to research more to find which glyphs aren't to exploit any entropy/equivalency of languages etc) - the glyph tests list from Fifield & Egelman's 2015 paper are not really fit for purpose and been on my ToDo for a long time, but good enough for this purpose - i.e we control the fonts per language