[CM] Worthwhile to explore realtime-friendly GC strategies for s7?

[bil at ccrma.Stanford.EDU]

In s7, cache misses happen almost entirely in the gc and eval functions.
In my informal tests, some of the standard benchmarks run ca 20% faster
on an AMD 3950 (3.5GHz) with 64 MB L3 cache than on an Intel i7-6700
(4.2 GHz) with 8MB L3 cache.  I'd like to try an AMD 5950 which I think
has a 256 MB cache, but chip prices right now are astronomical.
The benchmark compiler.scm, for example, runs in 23 secs on the
small cache and 16.5 secs on the large, even though the large cache cpu
is apparently slower.

You might also try a smaller heap, or use float or int vectors, etc.