I have the same code, written using win32com and xlrd. xlrd preforms the algorithm in less than a second, while win32com takes minutes.

Here is the win32com:

def makeDict(ws):
"""makes dict with key as header name, 
   value as tuple of column begin and column end (inclusive)"""
wsHeaders = {} # key is header name, value is column begin and end inclusive
for cnum in xrange(9, find_last_col(ws)):
    if ws.Cells(7, cnum).Value:
        wsHeaders[str(ws.Cells(7, cnum).Value)] = (cnum, find_last_col(ws))
        for cend in xrange(cnum + 1, find_last_col(ws)): #finds end column
            if ws.Cells(7, cend).Value:
                wsHeaders[str(ws.Cells(7, cnum).Value)] = (cnum, cend - 1)
                break
return wsHeaders

And the xlrd

def makeDict(ws):
"""makes dict with key as header name, 
   value as tuple of column begin and column end (inclusive)"""
wsHeaders = {} # key is header name, value is column begin and end inclusive
for cnum in xrange(8, ws.ncols):
    if ws.cell_value(6, cnum):
        wsHeaders[str(ws.cell_value(6, cnum))] = (cnum, ws.ncols)
        for cend in xrange(cnum + 1, ws.ncols):#finds end column
            if ws.cell_value(6, cend):
                wsHeaders[str(ws.cell_value(6, cnum))] = (cnum, cend - 1)
                break
return wsHeaders