On one hand, this tends to offer a slightly stronger assurance against Linux-level security faults while also enabling the use of non-Linux stacks (such as BSD or Solaris or - God forbid - Windows, along with just-enough-OS (or no OS whatsoever)). Proper virtualization like this offers another layer of security, and it's generally perceived to be a stronger one.
On the other hand, the security benefits provided on an OS level (since now even an OS-level compromise won't affect the security of the whole system, at least not immediately) are now shunted over to the hypervisor. Additionally, the fuller virtualization incurs a slight performance penalty in some cases, and certainly includes the overhead of running the VM.
On the third hand, bare-metal hypervisors tend to be very similar to microkernels in terms of technical simplicity and compactness, thus gaining many of the inherent security/auditing advantages of a microkernel over a monolithic kernel. Additionally, in many (arguably most) environments, the slight degradation of performance (which isn't even guaranteed, mind you) is often much more tolerable than the risk of an OS-level bug compromising whole hosts, even if the risk of hypervisor-level bugs still exists.