It's very easy to send an e-mail, and, consequentially, also simple to by mistake send something by mistake to someone. Some organizations are even required by regulations to perform content filtering of outgoing messages. With increased e-mail complexity comes increased difficulty to homogeneously enforce security policies.
By specifying your policies as content filters (or DLP patterns as they are sometimes referred to as) at the same central gateway where you perform outbound anti-spam, rate controls and message signing you get a clean, simple security implementation. Because attachments are traversed, archives (such as ZIP) is extracted and data is normalized to text, it's simple to write content filters.