The beauty of split tunneling is that your enterprise doesn't need to provide the general Internet access point for a VPN user. With split tunneling, the VPN client automatically determines whether a network location is accessible through the virtual private network and, if it is not, passes it directly through the network connection. If users send only 10% of their traffic to your corporate network, you're letting their current access provider handle the other 90% of the load.
On the other hand, split tunneling may leave users with a false sense of security. If they're following instructions to "connect to the VPN from the road," employees may believe that all of their traffic, including their personal email and Web browsing data, is encrypted by the VPN. They may not realize that the traffic is open to interception on the local network.
From a compliance perspective, PCI DSS doesn't make any statements about split tunneling. I believe you could construct a rational argument for either approach during a compliance audit. Split tunneling wouldn't really reduce the risk of malware infecting your corporate network. If it's present on a machine before it connects to the VPN, it will still be present on that machine when it connects, regardless of your VPN tunneling strategy. If you wish to ensure that systems connecting to your VPN are free of malware, I'd recommend investigating the use of network admission control (NAC) technology.
This was first published in February 2008