I don't believe I have it backwards, since RFC1191, the one you quoted,
states:
"A host doing PMTU Discovery must obey the rule that it not send IP
datagrams larger than 576 octets unless it has permission from the
receiver." So really, unless it get's permission to send larger packets from
the receiver, it'll send them at less than 576 bytes.
I imagine this'll all change though whenever IPv6 comes about since the min.
MTU size will be (approx.) 1200. As another poster stated, IPv6 requires
PMTU, so this could become a nasty little problem.
Thanks for the rest of the info on the HOW as I am not quite sure either and
your scenarios make sense..
Marc
I don't believe I have it backwards, since RFC1191, the one you quoted,
states:
Oh.. What you originally said was:
off ICMP routing. And no I do not believe it is used across the internet,
and if it does, it is probably hindering performance since it's probably
using a lower mtu than is allowed, such as 576 or smaller. It would also
Which I read as "and if it (PMTUd) *is* used, it is probably hindering
performance". If what you meant was "if it isn't used, you're stuck with
576, which hinders performance", that that is quite correct...
I imagine this'll all change though whenever IPv6 comes about since the min.
MTU size will be (approx.) 1200. As another poster stated, IPv6 requires
PMTU, so this could become a nasty little problem.
That other poster was me too... Unless I'm confused... 
Hopefully we'll get lucky and the clue level required to deploy IPv6 will be
high enough that people will get the obvious stuff right.