While it has done this for awhile I noticed a lot of likely un-needed disk I/O in Spark if spellcheck is enabled. Anytime within any chat window any character is generated or removed (client side input fields) there is a read request called to .jmyspell\en_US.idx and .jmyspell\en_US.dic
To keep with working with multiple languages this probably should be loaded into memory for performance.