The default load factor here was inherited from CRuby, which also
used a load factor of 5 for its "chained bucket" implementation of
Hash. Since then, they have moved on to a more space and cache-
efficient open-addressing implementation.

In jruby#9113, a user reported that large Hash initialization
regressed in performance relative to JRuby 1.7 and CRuby 3.4, and
this high load factor was found to be a big part of that.

A load factor of five means a saturated Hash may do 5x as much work
to find a given entry, because up to five chained entries must be
searched in a given bucket. Reducing this to a load factor of one
may be too aggressive, so we use a load factor of two here to cut
the maximum number of extra hops down to one.

Given 32-bit compressed OOPS and non-Liliput, we can do some math
for the size increase:

A 1000-element Hash at load factor = 5 consumes 800 bytes for the
bucket array and about 40_000 bytes for the entry objects for a
total of 40_000 bytes (not including the Hash object itself).

The same Hash at load factor = 2 consumes 2000 bytes for the
bucket array, an extra 1200 bytes or 2.9% size increase.