The ST3 engine is included within ST4 for compatibility with the older instruments. I did notice that if you are using the older ST3 based libraries in ST4, they are less CPU hungry. So if instruments are single threaded, then separate instances would be necessary to share the load. This seems to be more of an issue with the new ST4 library. I recall Noel mentioning somewhere that load balancing only applies to effects rather than instruments.
Includes over 600 GB of samples and over 18,000 instrument presets. VST2 or VST3 doesn't seem to make any difference in this regard, as well as setting plug-in load balancing on/off in Cakewalk preferences. The complete sound and groove workstation. With my old dual core at 3.4Ghz I only got about 3-4 instruments per instance.
When that core gets over 50% on that single thread is when I start hearing noise. If I run a CPU monitor I can see one core (of 6) doing all of the heavy lifting. With my current system, Intel i5-9600K at 4.6Ghz, I start getting breakups (pops & crackles) with 8 instruments loaded in a single instance of ST4 using ASIO at 256 buffers (24/48000). In practice, if you have 10 instruments in your project then few instances of ST4 with 2-3 instruments per each should handle it better than a single ST4 instance with all 10 instruments.