Does the software actually crash, or does it just hang for a long time? I've used CP to make image sets bigger than 35K, and it can take several hours (or even overnight) to finish loading them all, but it can in fact be done, at least on my machine.
I think what you should do is load a small subset of your data (maybe a couple of wells each from a couple of plates) into CP in a brand new pipeline, then play around with the Metadata and NamesAndTypes modules to figure out how you want to do metadata extraction of well name, plate name, channel name, etc, and how you want to match the channels together in NamesAndTypes. Once you're happy with it and it works across all your images in your small subset, save that pipeline and export the CSV of the image set (File -> Export -> Image Set Listing). From this point, you have two options:
1) See if you can now load the full image set into CellProfiler (it'll probably take many hours!) and run the same export command to get the full image set listing.
2) Use the CSV you have from your small subset experiment as a template of what these CSVs should look like; you can now write a script to read the image and/or folder names and spit out a CSV in the correct format in your favorite scripting language- almost any scripting language should be able to do this. This is more work up front, but if you'll be doing many large experiments like this it will almost certainly save you time in the long run.
Once you have done either 1) or 2), you'll have a CSV you can use for LoadData for your "main" pipeline that you're doing to measure your cells.