It seems to me that if we go for a strict non-overcommit strategy,
then the amount of swap that we need to allocate needs to really
be boosted by a huge number, and we'll never be able to get a very
high usage of said swap.

Specifically, what do we do in the situation of COW?  Especially
COW when a write is pretty infrequent, say when a process mmap()'s
in libc.  In this case, libc is mapped in COW (private, I think)
so that if you run a debugger on it the text pages can be modified,
so we go from a situation that generally requires no swap to (at
least on my alpha) requiring an extra Meg of swap for almost every
process.  Not to mention the X libraries, which are pretty large,
with a good few of them running at around 2 Meg.  If one is running
KDE, libqt.so on the alpha is around 4 Meg, and be linked into all
KDE apps.  Noting that to get KDE going one has to have about 8
processes started in one's .xsession, that comes to 32 Meg of swap
wasted on only libqt.so.

 == Roland Dowdeswell
 == http://www.imrryr.org/~elric/