These criteria are not intended to cause a subscriber to subnet Class C networks. If the subscriber's network is divided into logically distinct LANs across which it would be difficult to use the given number of Class C network numbers, the above criteria may apply on a per-LAN basis. For example, if a subscriber has 600 hosts equally divided across ten Ethernets, the allocation to that subscriber would be ten Class C network numbers; one for each Ethernet. Exceptions from the stated criteria would be determined on a case-by-case basis.
Oh, well... If some of you remember the form I sent to you as input to the common form, I specifically stated that even a class C network can be subnetted, and that such a strategy could be useful to conserve some address space. From the above, I read that this is not the intention anymore. Can someone explain why? (This is a section where the new version differs a bit from RFC 1366.) - Havard