All languages have redundancy; it serves as a form of forward error correction. I recall some sort of study (sorry, no link) finding that information density was (by some metric) similar across different languages. So it's not just inefficiency; it's part of the language's necessary inefficiency.
Of course, some forms of redundancy would be better than others--in cases where people aren't sure which particle to use, it probably isn't doing much good. However, language evolution is able to achieve some optimizations, and I suspect the particles people know tend to be the most important. For example, the many sushi or shellfish particles might sound particularly silly, but if you're in those industries, maybe they are helpful in maintaining important distinctions in a noisy kitchen/market, or in written records. If you're a customer you probably don't know them, and you don't need to.
Epistemic status: Wild-ass guessing from my armchair.
Of course, some forms of redundancy would be better than others--in cases where people aren't sure which particle to use, it probably isn't doing much good. However, language evolution is able to achieve some optimizations, and I suspect the particles people know tend to be the most important. For example, the many sushi or shellfish particles might sound particularly silly, but if you're in those industries, maybe they are helpful in maintaining important distinctions in a noisy kitchen/market, or in written records. If you're a customer you probably don't know them, and you don't need to.
Epistemic status: Wild-ass guessing from my armchair.