f688b9112d
* Use a buffered PRNG, pulling the PRNG data off a larger precalculated
buffer, rather than the underlying PRNG's (likely small) one, which in
turn reduces the frequency of recalcing.
* More tuning to reduce temporary allocation churn