So you don't expect anyone to ever test their modules against overloaded input? I disagree that comparisons should default to being x-ray comparisons just because they are part of a test module. I'm more likely to test that the output from two invocations produce the same apparent output than that they result in identical internal structures. The former is an important aspect of a module, while the latter shouldn't really matter.
Update: In a private message, diotalevi notes that he meant that the test should usually be normal but that there should be a way to do the x-ray test. My reponse is that to do the x-ray test you temporarilly unbless the overloaded items, but I'm shocked to see that Scalar::Util doesn't have unbless(). There is a non-core module, Acme::Damn, but that nearly doesn't solve the problem (being non-core and also XS and further Acme::). But I suspect that temporarilly blessing into a non-overloaded package would be a "fix", but I haven't tested such.