All of those, and probably more. You don't mention storage IO performance, for example, and that's usually important.
You need to measure performance of your workload(s) if you can, and you absolutely need to be measuring system performance metrics, so that when the former takes a hit, you can figure out the cause, which might be a bottleneck, which you will then need to fix or workaround.
So it's entirely dependent on your workload(s). If you know their performance limiters on this architecture very well, then maybe you can get away with only measuring a few performance metrics. But I'd advise against that, and measuring/instrumenting as much as you can.
Again: that's both the system metrics, and some objective external performance indicator of your apps! (and internal performance, if you can instrument that.) And this is in no way specific to VCE vBlock, or even virtualization in general. This really applies to any systems.