The file (and probably time) comes from a crazy number of these:

 <sheetData>
...
   <row r="51" spans="1:256" ht="19.85" customHeight="1">
      <c r="A51" s="6" t="str">
        <v>2.2.2.  initiative</v>
      </c>
      <c r="B51" s="6"/>
      <c r="C51" s="6"/>
      <c r="D51" s="6"/>
      <c r="E51" s="6"/>
      <c r="F51" s="6"/>
      <c r="G51" s="6"/>
      <c r="H51" s="1"/>
      <c r="I51" s="1"/>
      <c r="J51" s="1"/>
      <c r="K51" s="1"/>
      <c r="L51" s="1"/>
...

Specifically:

# egrep -c '<c r="[A-Z]+[0-9]+" s="1"/>' /tmp/ddd/xl/worksheets/sheet1.xml 
16768348

That's a lot of empty cells with style #1!

Plan of action:

1. Teach xlsx_write_col and xlsx_write_cols to write the most common
   style as the "style" attribute.

2. Teach xlsx_write_cells to only write existing sheets.

3. Make new xlsx_write_empty_cells_styles and have it create records only
   for non-existing cells with styles that are not the col's most common.

I think we have all the pieces to make this happen.

This will also fix bug 662058.

Created attachment 225986 [details] [review]
Partial patch

This reduces file size by three orders of magnitude and speeds things up a
good deal.

Created attachment 225997 [details] [review]
Updated patch

This also skips boring rows fairly fast.

> fairly fast

Fast enough to make the file from bug 662058 save instantly.

*** Bug 662058 has been marked as a duplicate of this bug. ***

This problem has been fixed in the development version. The fix will be available in the next major software release. Thank you for your bug report.