In my opinion it is a natural sequence of events:
 a demand of pipelining, error handling, and so on, is appearing;
 to answer the the demand people design extensions to xslt (functions and
instructions);
 these extensions are discussed and standardized;
 XProc is appearing as something based on xslt.

Is XProc so really required?

Why should XProc be designed rather than extending (providing new api for)
xslt/xquery to perform pipeline processing?

Are xslt/xquery less declarative than XProc?

I wouldn't probably asked such questions if XProc were already well
established with implementations available.

--
Vladimir Nesterovsky