Consider these things about Japanese phonetics (and forgive me if I am mixing up phonetics with phonology, phonemics, or phonotactics). There are five vowel sounds that combine with 10-odd consonant sounds to form the moras (basically syllables) that the kana represent. Four of those consonants (k, s, t, h) are voiceless (meaning there is no acoustic resonance involved in producing them), and have voiced versions, represented by the kana with two small lines in the upper right corner.

-Two of the vowels become voiceless (they are sort of whispered) when placed between two voiceless consonants, or after a voiceless consonant at the end of an utterance. These vowels are i and u. Hence, desu usually sounding like "dess."

-Out of the four voiceless consonants, three of them have allophones (different sounds) when in front of i and/or u. Instead of si there is shi, instead of ti and tu there is chi and tsu, and instead of hu there is fu (in fact not quite fu, but something close. Instead of the lower lip touching the upper teeth, the lips are brought near, and air is simply blown through them. But close enough.)

For a while I thought the first of these two aspects was simply a result of quick speech, and the second was just one of those weird things in language that you accept and move on. But language is so much more rational than we are often led to believe! I've just realized the clear relationship here. If all the voiceless i and u consonants were the same, it would be difficult to tell them apart when the vowels were devoiced. For example siki and suki sound quite similar when they are compressed to effectively both be ski, but when siki becomes shki there's no problem. The same goes for hi and fu. Chi and tsu provide a small problem if you're being strict, because it isn't necessary to change both of them from ti and tu. I think the answer here is that ti is simply difficult to say and moves naturally toward chi.
Ki and ku are more of a problem, with no allophones. The best explanation I can offer here is that it's simply not as difficult to tell those apart, for example in kiku 'to hear, to ask.' There's just something about the formation of k that requires clearer vowels, and doesn't allow them to be devoiced as much as with the other consonants.
Having done absolutely no research into this, it would be irresponsible to say there must be a direct causal relationship from the devoicing to the allophones, even though it seems like the simplest story. Language isn't that rational! Things get all mixed up, causes and effects are inextricably intertwined.

Meanwhile, I've begun investigations into learning Welsh, as I promised Gruff Rhys I would do starting at the end of the year. It seems that just about every source available, other than a few super-expensive CD sets, is written by a guy named Gareth King. He has produced such wonders as "Basic Welsh", "Modern Welsh", "Colloquial Welsh", "Intermediate Welsh", and "Pocket Modern Welsh" (Dictionary). Colloquial Welsh was co-written by Gary King. The two editorial reviews of it on amazon are from Alan King. There's something suspicious here, I can't quite put my finger on it...