One or two people have written to me to suggest that I discuss the English pronunciation of tsunami. I have not done so, because the matter has already been extensively aired on Language Log, including here.
OK, then: let’s agree that if we articulate very carefully we can reproduce in English the Japanese affricate ts (which is the allophone of /t/ used before u), giving tsuˈnɑːmi. But once we have got into our stride we’re likely to simplify the initial affricate to a plain fricative, suˈnɑːmi.
(Fun fact: in the local speech of Kōchi 高知, on the island of Shikoku, this allophone is not used, and people say tunami.)
The same sort of thing happens in other borrowed words that have initial ts in the donor language. So tsetse (fly), from Tswana, becomes English ˈ(t)setsi. However with tsar (from Russian царь tsarʲ) and Zeitgeist (German pronunciation ˈtsaitgaist) the English outcome tends to be a voiced fricative, thus ˈzɑː, ˈzaɪtɡaɪst. In the latter word this is easily explained as a spelling pronunciation giving the letter Z its usual English value, and I suppose tsar may be similarly influenced by the alternative and older spelling czar, which the OED dates from 1549.
This voicing uncertainty made me think about other cases where we have borrowed from a foreign language words beginning with a consonant plus s. Notably, these are the Greek-derived words spelt in English with ps- and x-.
It is clear that words with Greek initial psi (Ψ, ψ, ps) end up in English with a voiceless s, while those with initial xi (Ξ, ξ, ks) end up with z. So on the one hand we have psalm, pseudonym, psoriasis, psychology etc with s-, but Xerxes, xenon, xenophobia, xylophone with z-.
So the bilabial produces a voiceless outcome, the velar produces a voiced outcome, and the alveolar can go either way. I do not know why.