> It would be nice if a particular XSLT processor implements lazily the
> index-of() function,

Saxon does so.
>
> Of course, exactly the same reasoning applies to the implementation of
> <xsl:key> and the key() function, if a smart enough XSLT processor
> allows the key() function to operate on a partially-built index, and
> when a solution is found the building of the index is halted.

That's certainly theoretically possible, though it's considerably harder to
achieve if only because of the multi-threading complications.

Michael Kay
Saxonica