I like what Liam is recommending better than the other solutions
(including my own), but I think it is addressing a slightly different
problem than what OP posted (at least, slightly different than what I
perceive the OP posted -- I could be off. :-), although it may well
be the problem OP really *has*.

Crism's algorithm addresses the same problem mine does. A quick test
says Terry's does, too, although it seems a bit complicated to me.[1]
One of the things I like about my "append delimiters,
normalize-space(), remove delimiters" approach is that because it
uses only simple XPath 1.0 functions, it can be used in XSLT 1.0 and
I imagine is very fast. (I think Crism's approach probably has
similar advantages.)

The following is a complete XSLT 1.0 program that regularizes
whitespace as I think the OP was requesting.
--------- start ---------
<?xml version="1.0" encoding="UTF-8"?>
<xsl:stylesheet xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
version="1.0">

  <xsl:template match="@*|node()">
    <xsl:copy>
      <xsl:apply-templates select="@*|node()"/>
    </xsl:copy>
  </xsl:template>

  <xsl:template match="text()" priority="1">
    <xsl:variable name="intermediate" select="concat('b ', ., 'b ')"/>
    <xsl:variable name="semifinal" select="normalize-space( $intermediate
)"/>
    <xsl:value-of select="substring( $semifinal, 2, string-length( $semifinal
) -2 )"/>
  </xsl:template>

</xsl:stylesheet>
--------- end ---------

Note
----
[1] Question -- Terry uses <copy-of> where I use <value-of>. In both
    cases the select= holds a (variable that is a) string. Is there
    an advantage of one over the other?