The SQLite version is reputed to be much faster since the database allows it to more easily determine which schedules need to be downloaded. (I've not tried the non-SQL version).
I have a fork of the SQLite version (tv_grab_az_sdjson_sqlite) which adds a few features such as artwork and optional better caching which makes it much faster overall if you have huge numbers of channels, but it requires a lot more effort to setup. So I'd probably stick with the supported SQLite version first and only switch once your comfortable and really want extra features.
All xmltv grabbers actually generate a file in a common format and that common file format is then read by the tvheadend/mythtv. So they don't read the SQLite database themselves. However, the xmltv file format allows optional sections, such as artwork, episode numbers, etc., and not all grabbers populate all the information.
You can share the grabber config with both tvh and mythtv but it's a bit tricky since they use different user ids (permissions). However, assuming your Internet is unlimited, then the easiest way is to just configure it under both tvheadend and mythtv with the same SD login details and different directories/databases and just let them both fetch the listings information independently in to their own separate SQLite databases.
Though, you seem to suggest your mythtv is running on a separate box, so yes, just configure it with the same SD details as your Pi box.
