The problem is, that 10k to 15k of files may very well contain 15k-30k keywords to be saved that could be searched for with complete substring matching. Not even google.com does substring matching. With 30k keywords, you will easily receive a few hundred search queries per minute and matching each search query against a 30k keywords is very cpu intensive.

As far as I can tell, LimeWire's search algorithm is already pretty optimized. And it won't be easy to increase performance so you can easily share your 15k files.