Here are your options. I know the #nochanges crowd will not like logic but please read through and say which version is most likely
You have 3 true options:
1) Small server sizes with many servers(OG Vanilla style). Which has higher operating costs(more servers running), higher initial costs(more hardware purchased) and will never see a true drop in operating costs because each server will have X dedicated resources. This is not scaleable and in the IT world and Blizzard word this is the biggest key to any operation. Does this layout allow for us to grow or shrink with demand.....and the answer is NO!!!!! This also has the issues of players becoming frustrated initially with high queue times as there will be strict limits on logons, etc.
2) Large server sizes with no Sharding. This has will have the same high operating costs and high initial costs but will have capabilities to allow for a drop in costs but these drops will take maintenance windows to preform as you will be resizing VM's, moving VM's from one cluster to another, etc. This will have a higher management cost and if the they estimate wrong they will be stuck waiting for the next maintenance window to increase server capacity. This solution once again is not overly scaleable as the management overhead is prohibitive. This will also have the issues of launch when you have 1000+ characters fighting over the same mob, leading to decreased retention rates.
3) Huge servers with Zone Sharding active(possibly even 1-2 servers per type: PVP, PVE, RP and RP-PVP). This has lower operating costs because everything is done in the container world of programming and "shards" or containers can be spun up automatically as needed and when they are no longer needed the container is shutdown are removed and these resources are returned to the pool. This makes this solution completely scaleable. As these servers share the database with live they will be able to share hardware with live at the same time, thus splitting hardware costs with live. This also has lower management because most of it is handled through automation. The only draw back is it does not completely appeal to the "no changes" crowd but this can also be mitigated with no sharding zones(major cities, RP favorites on RP servers, etc). This also removes the need for CRZ as there will be only a few realms to begin with.