SIP is designed for connecting endpoints (clients and proxies), but NOT for conveying content; SIP is NOT a content standard. Conveyed content could include VoIP, InstantMessenger, thermostat data, etc. SIP is "medium agnostic."
Each SIP user agent connects to a SIP "proxy" which performs several functions: Registration, proxy, and location service. SIP proxies can find other users, rather like a dynamic "white pages" database capable of linking each user Address of Record (AoR) to multiple network or device types. SIP can support call forwarding, caller and callee authentication, call transfer, transaction logging, access restriction, etc. (More info at http://www.cs.columbia.edu/sip/overview.html)
"If a SIP server can decide which of several media should be utilized at a given moment for presence-based interaction, SIP itself my become a killer application." Each SIP user has Address of Record (AoR), which may look like an e-mail address," but SIP is actually "address neutral," in that the address syntax is not defined. For example I might have a VoIP, video conference, AND instant messaging address like
sip:smith@sip-server.ku.edu
SIP-enabled clients would contact the specified SIP server to determine my current numeric (dotted quad) address and possibly recommendations for an acceptable content medium (text, voice, low-res video, high-res video, etc.) and suitable configuration parameters for endpoint devices.
SIP can also be used to convey "current presence" or some "indicator of availability" as content. This would be basically a standardized buddy list. The issue is not "whether to provide SIP," but "when". If you need whitepages now, use 323, but SIP is coming and you should be prepared for mixed 323 and SIP networks. SIP will provide an infrastructure for whatever apps are created for the mobile environment, especially with mixed endpoints.