64Gb of memory isn't all that much once you take into account FS cache with a decent sized index. I wouldn't recommend doing multi nodes on such a server - rather just do a bit of tuning and spread the I/O across multiple disks if you can.
I'm actually looking at doing option #3 but only because I have one very large machine (512Gb mem, 32 cores, etc) to work with. It's not what I would term as ideal but it's what I have to work with. This doc (Can I run multiple Elasticsearch nodes on the same machine?) does a great job of describing how to accomplish it.