Expensive XSLT2 - suggestions for improving?

The task is to remove duplicate text content before moving an XML file
into translation. After the translation, the former duplicate content
should be recreated.

<Doc>
<value oid="40068">Lasttrennschalter</value>
<value oid="40069">Umbau von N12 auf N4</value>
<value oid="4006a">Lasttrennschalter</value>
</Doc>

The third <value> should be empty because its content is identical to
the first, but we need a pointer to that first element to be able to
recreate the content after translation. Also, all original attributes
must stay unchanged. Therefore in each duplicate I insert an extra
attribute @refoid with the @oid of the source element. So I get this:

<Doc>
<value oid="40068">Lasttrennschalter</value>
<value oid="40069">Umbau von N12 auf N4</value>
<value oid="4006a" refoid="40068"/>
</Doc>

My XSL is very simple and works as intended, but it does not scale
very good, I guess because I look at preceding::value so many times:

<!-- Condenser: modify all duplicates -->
<xsl:template match="value[.=preceding::value]">
  <xsl:copy>
    <xsl:apply-templates select="@*"/>
    <xsl:attribute name="refoid"
      select="preceding::value[.=current()][last()]/@oid"/>
    <!-- skip content -->
  </xsl:copy>
</xsl:template>

<!-- pass-through all nodes and attributes -->
<xsl:template match="@*|node()">
  <xsl:copy>
    <xsl:apply-templates select="@*|node()"/>
  </xsl:copy>
</xsl:template>

I guess a clever constructed key could help a lot... any pointers are
very welcome!

--
_______________________________________________________________
Michael M|ller-Hillebrand: Dokumentations-Technologie
Adobe Certified Expert, FrameMaker
Lvsungen und Training, FrameScript, XML/XSL, Unicode
<http://cap-studio.de/> -- Tel. +49 (9131) 28747