I assuming you're not asking why Chrome doesn't do this ('cause the
proposed spec said it?), but rather why the spec does indeed require
this.
I vaguely recall that the restriction was suggested analogously to
handling of cookies, and is based on:
http://www.ietf.org/rfc/rfc2965.txt
which has pretty exactly the restrictive wording seen above. Search
the RFC for "contains one or more dots."
If these restrictions were not in place, one example of a Reduction Of
Service attack would be for a malicious party to claim (somehow) that
YourFavoriteSite.com has a few hundred dictionaries :-/. If these
were established in clients, then each HTTP request would have to list
all their hashes :-(. That in turn would potentially slow requests,
and require increased bandwidth when communicating with
YourFavoriteSite. I think this same issue is part of the reason to
restrict cookies.
YMMV... but that is some reasoning that comes to mind.
Jim