> unless Saxon performs some smart optimization. What do you think ?

even if you don't use a key saxon (perhaps  just in the -sa version, I
can't remember) may decide you use some //foo[@x='aa'] idiom often
enough to internally use a key anyway. I tend to think of for-each group
as a sane syntax added at xslt2 for the muenchian grouping idiom that
Dimitre showed, which works well, but only looks normal as its posted
once a week for over half a decade to this list:-). No idea if the
internal implementations are similar though but that's processor
specific detail...

.
Completely agree with David.

Not speaking about performance but about *style* and expressiveness,
there are some interesting things to note:

 2. No need to pass parameters using the "keys" approach-- most of
the work is being done as one atomic operation -- the one-time
creation of the index

 3. Some purists will comment that <for-each-group/> supports the
"more imperative-like", "pull" style of using <xsl:for-each/> as
contrasted with the more general and flexible push-style
<xsl:apply-templates/>

Threfore, one might conclude that the "keys" approach can be
implemented in a more compact and flexible way and the resulting
implementations can be more maintainable.

--
Cheers,
Dimitre Novatchev
---------------------------------------
Truly great madness cannot be achieved without significant intelligence.
---------------------------------------
To invent, you need a good imagination and a pile of junk
-------------------------------------
You've achieved success in your field when you don't know whether what
you're doing is work or play