> Haven't measured it [the performance benefits of keys] though.

I have. The performance benefits of using keys are substantial.

I have two versions of a program to join the <row> elements in <ARPT> with the
row elements in <ANAV> with the <row> elements in <BDRY>:

1. The first version uses an xsl:for-each within an xsl:for-each within an
xsl:for-each.

2. The second version uses keys.

On my large input XML document:

1. The first version took 738 seconds (12 minutes, 18 seconds)

2. The second version took 25 seconds.

The version using keys is 29 times faster.

/Roger