It's not an illusion but virtual memory management working as intended - with a little help from Apple's unified memory and their fast SSD implementation.
Virtual memory systems do allow you to over-subscribe memory. They deal with that by keeping a working set pages for active applications in memory. To make room for those pages, macOS resorts to removing unused pages, removing file system cache pages, paging applications and swapping applications (traditional OS methods) as well as compressing unused memory.
To macOS, a virtual machine's memory is somewhat opaque. It knows it's being referenced, but it doesn't know or care what's in it. An 8GB VM will be allocated as 8GB virtual address space. That doesn't mean it will take up 8GB of memory. Last I checked, Fusion does not wire a VM's memory footprint, so it's eligible to be paged in and out by macOS as it deems fit. It'll try to keep however much of the VM's memory footprint resident depending on how the VM is using it. macOS doesn't have to move the entire VM's memory footprint to the swap/page file, just the least recently used portions - unless over subscription gets *really* bad.
Where things get interesting is when you have over-subscribed memory too much and the actively running processes (including VMs) causes repeated page/swap in-out whenever a process/thread is scheduled to run. That is reported as high memory pressure and a drop off of performance in the VM.
As a side note, you sort of have dueling virtual memory management here. The VM thinks it has memory equivalent to what you set it for. It will manage virtual memory for the applications that it has running. macOS will manage the memory of the VM application at the system level. That gets interesting when you've oversubscribed the virtual machine's memory where the VM's OS is paging in and out, and you've oversubscribe the host OS memory so it has to page things in and out. Both operations are independent of each other.
This is where unified memory and the SSD on Apple silicon come into play. Those are much faster than in earlier Macs. The user sees less performance degradation because of the speed of the page in/out operations.
However, as in life, there is too much of a good thing. You can still over-subscribe VM memory to the point where the VMs performance is impacted. I remember this being referred to in the classic CS texts as "thrashing", where the system is spending more time managing memory in and out than actually doing work. And consider that constant paging out is writes to the SSD - which over time may impact the lifespan of the the SSD.