I just did commit 9cd7a95a2bb9ba9d5a6af022e5454c8ce78aee56 "[gdb/testsuite] Fix gdb.base/float128.exp with --with-mpfr=no", where we do:
...
+       if { $mpfr_supported == 0 } {
+          # If the host platform has native 128-bit float support (as is
+          # the case for some versions of s390 and powerpc), the
+          # "print large128" test should be passing, even without MPFR
+          # support.  So, in those cases we should have fail here rather than
+          # unsupported.  However, given that we don't have a way to readily
+          # test for this, we fall back to unsupported.
+          unsupported "$test (Missing MPFR support)"
+       } else {
...

In principle, gdb could make a statement through some maintenance command about what the expected host-side float point precision is. [ I'm sort of hoping there already is one, and I haven't found it yet. ]

Then, using this maintenance command we could make this test more accurate.