All but the most performance-demanding environments should be fine with the OS sharing spindles with indexes. There isn't that much "OS" I/O on an indexer, especially if there is enough memory to avoid swapping entirely.
But, I would suggest that if you have any machines acting as both search head and indexer (obviously, dedicated search heads are best ... but just in case), that you reserve some spindles to keep the indexes separate from the search dispatch directory. That might be a good example of OS+swap+Splunk product on one set of spindles, and indexes on another. That way, when a search is running, you don't have the same set of drives seeking to read from the indexes and also seeking to write the search results into dispatch.
