The runtimes are also written in “memory unsafe languages” (C++). The runtimes bring a whole lot more code than if you wrote own tailor made code, meant to do something specific, in “memory unsafe languages”.
This is not correct in my experience. I think it's more apt to say that security effort has mostly been spent around sandboxing and related technologies which is really an admission that there is no way to secure the JS VMs in themselves. The best engineers in the world can't do it.
Maybe that will change if they move to safer languages, but so far nobody has done that.
Therefore when you see an exposed unsandboxed VM, you instantly know it's critical issue.
Writing secure C++ is quite hard, even for the best engineers in the world. However, that absolutely does not mean that your handcrafted C++ code is more secure than JavaScript running a virtual machine that is written in C++. The sandbox exists as another layer of defense, not because the code is inherently more insecure. (Also, it's usually because JavaScript virtual machines evaluate untrusted input, which is something that has been shown to be notoriously difficult to secure against in general.)