Using plain hand-coded functions (fast-plural-rules) is more than 300x faster, than parsing the CLDR rules on the fly (plural-rules), as shown by a simple benchmark, while still keeping the library size smaller.
Intl.PluralRules has been implemented in the recent browser and Node.js versions. When creating a new Intl.PluralRules object for each cardinal computation, the performance is 2x worse or the same as when parsing the CLDR rules. When pre-creating a new Intl.PluralRules object and reusing it later, it showed 20-60x better performance than when parsing the CLDR rules. Using plain hand-coded functions is still 5-6x faster than a pre-created Intl.PluralRules object.
$ node perf
Getting a plural form...
using a simple coded rule x 3,013,201 ops/sec ±1.14% (90 runs sampled)
using a simple parsed rule x 21,257 ops/sec ±2.64% (84 runs sampled)
using a simple native rule x 12,030 ops/sec ±1.28% (89 runs sampled)
using a simple native cached rule x 499,103 ops/sec ±0.89% (89 runs sampled)
using an average coded rule x 2,083,520 ops/sec ±1.07% (91 runs sampled)
using an average parsed rule x 11,050 ops/sec ±1.67% (82 runs sampled)
using an average native rule x 11,340 ops/sec ±1.08% (89 runs sampled)
using an average native cached rule x 485,830 ops/sec ±1.05% (90 runs sampled)
using a complicated coded rule x 2,099,193 ops/sec ±0.81% (89 runs sampled)
using a complicated parsed rule x 8,004 ops/sec ±1.46% (90 runs sampled)
using a complicated native rule x 10,499 ops/sec ±1.63% (87 runs sampled)
using a complicated native cached rule x 428,312 ops/sec ±1.03% (86 runs sampled)