In the case of English, it's because of the knowledge of and prestige of Greek (and Latin). The higher more educated classes, who were traditionally taught classical languages, preferred to stick to the original declension, and they could spread this preference through grammars, dictionaries and schooling - but only to a certain degree. So it's a somewhat artificial phenomenon, and it tends to have only a limited spread outside of those who were taught Latin and Greek (so people do spontaneously say octopuses, and sometimes they try to make a classical plural that is nonetheless "wrong": octopi).

Similar stuff happens in some other European languages, I could list a few examples from my native Croatian where Latin grammar is supposed to be used, according to the wishes of some classical philologists, even when it clashes with all rules of native grammar (and, as expected, the latinate grammar is not used by anyone outside of a handful of academics).

At worst, such usage could be interpreted as social signalling of superiority (class + education). Obviously, nobody expects people to use "original purals" in cases such as gulag-gulagi, or wunderkind-wunderkinder, because Russian and German aren't all that sexy, even though German is much closer to English than Latin or Greek are.

But there are more natural cases of loaning inflectional morphemes, in communities with a very high degree of bilingualism. I'll admit I could only copy-paste some passages from books regarding this, I don't have much knowledge of languages outside of Slavic and English.

It's because -us is the Latin second declension nominative singular, and its nominative plural is -i. People educated enough to know the -us/-i pattern (cactus/cacti, alumnus/alumni, stimulus/stimuli) but unaware that "octopus" is Greek and has an irregular stem are likely to misapply the Latin pattern.