I was mostly referring to the costs to host, in the context of discussing what it would cost to maintain an MMO that already exists and is paid for.
If you're going to host in a virtualized environment then there's no real option but Datacenter licenses. The flip side is that Datacenter covers all the VMs on the same hardware which usually saves you some money. Whether CoH itself would be virtualized is a separate question. I don't think latency is as much of a problem as it used to be. Latency-sensitive stuff like VOIP systems are virtualized all the time now (I've done it).
One place I know for a fact CoX had scaling issues was in the database: I crashed the beta server when I tried to do something that ultimately required a lot of database activity (a zone-wide power grant). But compared to 2004, or even 2012, you could host the database server in an environment where the disks were all SSD, and on top of that most or all of the entire active databases were cached into memory. You combine that with memory-clustered synchronous writes and I don't think that would be a problem today. However, this is more a question of how you might make it better today than it was before, not whether it is possible to host something comparable to what existed before.
I don't think I specifically said I would host an MMO on cheap NAS storage. However, on that subject there are ways to make cheap NAS storage extremely fault tolerant in a virtual environment. You can continuously replicate them with ZFS snapshots. You can use disk replication like HAST or DRBD. You can also do some interesting things with dual port SCSI. You can even get zero-complexity redundancy by simply using two high performance NAS pods and provisioning your virtual machines to software mirror across them. Although I would not recommend that for ultra high disk performance windows use cases, because Windows software mirroring sucks. For high availability Windows with only moderate to high disk i/o requirements it works, and it works extremely well for Linux and BSD environments where software mirroring is extremely efficient.
Ultimately, the generic statement about how difficult or costly it would be to host an MMO environment has to be qualified by the scale of the workload. Of course, anyone can host an MMO environment if the simultaneous player count is three. It gets interesting if the count is three hundred. Its a real problem if its three thousand. When its thirty thousand, now you're talking about nitty-gritty details becoming important. I think somewhere between three hundred and three thousand the problem stops being trivial and starts requiring targeted solutions. Between three thousand and thirty thousand things get speculative without concrete system requirements. But I do have a datapoint, and that is that NCSoft was doing this in 2012 with circa 2010 hardware and it wasn't ridiculous in scale. A rack row of 2010 hardware is about two racks of 2016 hardware. Nothing NCSoft was doing with CoX in 2012 would be immune to Moore's Law that I'm aware of. And that places strong limits on the costs to replicate today.