As you can imagine, watching out for duplicates and performing the merge adds quite some extra processing time. Depending on the data type, merging conversions are about 50% to 300% slower than simple conversions. It's therefore recommended to switch merging off when you convert single files. That way you also don't risk problems with homonyms.
There shouldn't be any issues with memory, though. Any size of termbase can be merged, stopped only by the limits of MultiTerm if you convert to sdltb.
Created with the Personal Edition of HelpNDoc: What is a Help Authoring tool?