Indexes are kept in a tiered structure, in what sort of amounts to "generations" of data, e.g. long term stable index entries are bunched in larger, older generations, newer entries in small, young ones, This is done to keep performance of tracking / updates as high as possible. Every so often, when some generation grows over a calculated threshold, it needs to be merged into the next older generation. So the interval is directly dependent on the number of already existing entries, the frequency and pattern of updates, and can't be specified in absolute terms like "every x days".
Another thing - apart from merges - that can lock up significant disk space in the old Lucene system, is the result of programming errors in Alfresco customisations. When search results are not properly closed in Java code it can leave index files alive even after they should have been merged, which also won't be cleaned up until Alfresco is stopped (blocked by active file handles). In my experience, this is actually the most likely cause of high disk usage - not index merges. Unfortunately, this may be hard to fix when those incorrect code usages are in 3rd party customisations.
