If you have not done so already, I would suggest you use Devel::NYTProf to identify where all the slow points are in your code. It will save a lot of guesswork, and there might be other parts of your codebase you had not thought of as causing bottlenecks.

In terms of the example code you gave, I have not used Win32::OLE, but I/O can often be a bottleneck in code. Is it possible to avoid repeated reading and writing of the file? i.e., read as much as you can in batch mode (a data range), update it in perl, and then write it back in one go? That said, you should run your code through Devel::NYTProf to make sure this is the case before you do any editing, and keep track of the timing changes to see if it makes a meaningful difference.