Indeed. But the cool thing, I guess, is the big number of clusters in the calculator the similarity rate and map are based on, which serves the purpose of these tools, rather than being informative per se, in isolation. In fact, the tools work like an Oracle more or less, and they have a decent use for comparisons, making more sense when analysed as a whole. Plus, I guess a high rate tends to evidence, comparatively, a real high similarity. It's just that there must be a margin of error for them, naturally. Still, it's a nice reference, as we can observe empirically. Not perfect, of course.
As for Minoans etc., the similarity rate uses a calculator with modern references, meaning the DNA of ancient samples are categorized in clusters based on modern individuals, and it doesn't matter the "distance" of each fit. So it's inverted. The purpose of calculators is to find how the individual is mixed, either using contemporary references or ancient. But ok.
It seems K36, as all of its kind, calculates which cluster is the closest to a certain segment, and once it finds one, all the others are despised (zero fit). What matters are the overlaps of the results themselves, or if you prefer, the differences (100 - <differences>). So, if you get a relatively good fit of 5% of certain cluster, and the ancient sample gets, say, a bad fit of 10% (but still a fit; it will be forcibly categorized anyway), you'd get 5% more of "similarity" rate. As an extreme example, imagine a Denisovan against these tools. Let's hypothesize he gets some relatively high % related to certain Austronesians. The result is that the similarity rate between these Austranesians and a Denisovan could be higher than between the former and, say, the Sardinians(?).
In short, the clusters are mutually exclusive. They become more informative when well chosen, as the reference samples.
As you said in another thread, the results are not a given truth. A last example, practical, and not so extreme, are the Basques, a somewhat drifted pop. I'm not sure they get more Neolithic DNA than North Italians in admixture tests, for example. Probably not. However, Sardinians and Basques, "more than any other populations in Europe", would be those who best "preserved the 'original' widespread early Neolithic population component". Then I think we could say Basques are the second more similar to early farmers genetically, according to f-statistics. But not necessarily according to admixture.
@Angela
I just checked my K36 results. I see some inconsistence just in Eastern_Euro and West_Caucasian clusters. Mine are higher than the sum of my parents'. That's using just 23andMe data. This fact could in theory result in a higher score than the sum of my parents', yes. I haven't checked if it happened though.
Btw, the numbers in your similarity map above are a bit offset. It's likely a browser issue. Perhaps it has something to do with the zoom? Anyway, again, it's somewhat similar to mine, for obvious reasons.