>
> For point of argument, let's say that an attribute that is an ID or Key will likely be indexed (xsl:key or some hash table or btree or list or whatever) by any process complex enough that we really care about memory usage (given that a process may be short on memory because we need expedition.)
>
You're obviously right: all performance measurements suffer from the problem that data for one workload doesn't necessarily extrapolate to another.
Personally I would expect that for most documents, indexes aren't going to make a big difference (unless you're doing full text indexing), but I can't quantify "most", so that's a fairly meaningless statement.
Michael Kay
Saxonica
