The XServe is in an interesting spot relative to the x86 transition. On one hand, a lot of XServes are basically running whatever services Apple ships with it--fileserver, print server, webserver, directory services, etc. For those users, you could switch to x86 today and nobody would know the difference. On the other hand, for these users, if the PPC is performing equal to or better than whatever x86 chip would replace it, there's also zero reason to change.
On the other hand, there are people with specialized clustering or single-purpose apps. If these apps are already set up to take advantage of the G5's vector unit, or just designed by companies that are slow moving, folks aren't going to be too anxious to switch. Furthermore, if, as some special-purpose high-end benchmarks have shown, the G5 is on the leading edge of performance for certain specific kinds of apps, there's not much reason to change over until that's no longer true, as application compatibility isn't a big deal.
For this reason, I'd guess the XServe will be the last of Apple's machines to go x86. If IBM can offer Power-based servers for specific market segments, certainly Apple could as well.