Alex Shinn wrote:

> First, with 3-bit tags the max fixnum is 2^28-1 = 268435455,
> half your value... and half the memory, which already means
> we have more than enough remaining memory to implement
> something like memcached.

I said "the largest positive fixnum is likely to be 536870911."
If you survey implementations that use 3-bit tags, you'll find
that most devote two tags to fixnums, doubling the fixnum range.
If your reasoning applies only to systems that don't do that,
then your argument applies only to systems whose performance is
likely to be unsatisfactory on other grounds.

> The keys could be fixnums.  Since we can use negative fixnums
> we already have more possible keys than the maximum number
> of buckets.

That's a consequence of SRFI 126's arbitrary restriction of hash
values to non-negative integers.