Here's an example scenario of where VoodooShield would fail.
John Doe is browsing the web and he gets a notification about an e-mail to do with his work-place, so he goes back to the Outlook tab and checks out the e-mail; the e-mail has been spoofed to appear as though it was coming from his manager, and has been sent to several other colleagues as well.
John decides to download the attachment from the e-mail - he's using an Anti-Virus product with real-time protection enabled as well as VoodooShield on the side. The Anti-Virus product scans the downloaded attachment, but the downloaded attachment was specifically crafted to exploit a previously unknown vulnerability in the Anti-Virus product's scanner engine.
As the Anti-Virus product's scanner engine is scanning the contents of the attachment after mapping the file into memory, it runs into a bug because of a logic flaw in the engine. However, this flaw has been exploited in a way that allows shell-code planted within the file being scanned to be executed under the context of the Windows Service which is scanning the data.
Now, malicious shell-code can be executed under the context of something which is trusted and already in-memory (and thus, VoodooShield is not going to be blocking it, nor will it have any idea that process X has been compromised).
Obviously, even if a trusted running process were to start spawning other programs, that may trigger VoodooShield depending on the configuration (not guaranteed because it's configuration-dependent). Therefore, you'd preferably need to avoid doing this at all costs to prevent potential intervention from VoodooShield.
The example scenario is not something that a normal user should care about. Even for targeted attacks, such is incredibly rare because most of the time, it's found out by security researchers who report it and get it patched before a malicious actor can use it.
Microsoft recently added a sandbox container to Windows Defender (currently optional) to try and restrict what malicious code could do if it managed to compromise a user-mode Windows Defender process, like as a result of things like exploiting a flaw in the scanner engine.